Quartz
중도 성향
IT/기술
Apollo and Blackstone are arranging $36 billion to buy Google AI chips for Anthropic
The firms plan to buy Google tensor processing units through a special-purpose vehicle and lease them to the AI startup
🇺🇸 미국 · "PROCESSING" · 총 41건
필터 보기현재 지수
50.0
0 = 부정 우세
50 = 중립
100 = 긍정 우세
최근 7일 기준 11,953건을 분석한 결과, 뉴스 심리지수는 50.0(균형)입니다. 긍정 1건(0.0%)·중립 11,951건(100.0%)·부정 1건(0.0%)이며, 중립 비중이 뚜렷하게 높습니다. 성향 지수는 종합 19.1(중도 균형)입니다.
The firms plan to buy Google tensor processing units through a special-purpose vehicle and lease them to the AI startup
On Thursday’s broadcast of the Fox News Channel’s “Fox & Friends,” DHS Secretary Markwayne Mullin said that he isn’t planning to pull customs processing from all sanctuary cities, just in situations like in Newark where local law enforcement won’t protect The post Mullin: We’re Not Going to Pull Customs from All Sanctuaries, Just in Situations Like Newark appeared first on Breitbart.
Dell has gone from being a sleepy legacy tech company to a high-growth AI story, assembling servers packed with graphics processing units.
U.S. Citizenship and Immigration Services is tightening the rules governing how immigrants already in the United States can obtain green cards, a change that could require more applicants to complete the process abroad through consular processing. Under a policy memorandum issued last week, the agency said it will generally reserve “in-country” green card approvals, known […]
Homeland Security Secretary Markwayne Mullin said the Trump administration is drafting plans to stop processing foreign travelers at airports located in sanctuary cities, but the possible change has some former officials questioning the political implications. Mullin revealed on “Hannity” Tuesday night that officials are “currently drawing up plans” to pull Customs officers from international airports ...
Homeland Security Secretary Markwayne Mullin said the Trump administration is considering pulling international flight processing in "sanctuary cities."
Existing generative AI models are built on batch processing: You give the system instructions; it runs computations; then spits back the results. Now comes a Silicon Valley startup that says it can produce gen-AI video (and other outputs) in real time — a potentially groundbreaking advance: San Francisco-based Reactor, co-founded by Alberto Taiuti and Bryce […]
Gov. Gavin Newsom signed legislation that tries to stop outside officials from taking ballots or getting involved in ballot processing.
The Russian government has reached the final stages of implementing a comprehensive ban on diesel and aviation fuel (kerosene) exports following a devastating wave of Ukrainian long-range drone strikes that have knocked out a quarter of the nation’s total oil refining capacity, effectively paralyzing critical energy infrastructure across central Russia. Key energy hubs hit in recent weeks include the Ryazan, Moscow, Kirishi and NORSI (Nizhny Novgorod) refineries processing roughly 238,000 tons per day (83 million metric tons per year), accounting…
China’s top offshore crude oil and natural gas producer, CNOOC, on Wednesday announced the launch of full production from phase one development of the Kenli 10-2 oilfield in the south Bohai Sea. The Kenli 10-2 oilfield cluster is China’s largest offshore shallow-layer lithological oilfield and will be developed in phases. Phase One now yields more than 20,500 barrels per day (bpd) of crude oil after full production started. The phase one development includes a new central processing platform, two unmanned wellhead platforms, and 79…
The Trump administration is making plans to pull Customs officers from international airports as a punishment for sanctuary cities that refuse to play ball with federal immigration authorities. Homeland Security Secretary Markwayne Mullin revealed on “Hannity” Tuesday night that officials are “currently drawing up plans” to stop the processing of international travelers into sanctuary airports. ...
The United States declared several national emergencies targeting its critical mineral supply chain vulnerabilities. Through executive order 14156 and related directives, the Trump administration has framed overreliance on Chinese-dominated supply chains as a direct risk to national security and economic growth. Beijing’s control over both production and processing of key materials critical to defense, coupled […]
The damaged tank at Nippon Dynawave Packaging Co. held approximately 900,000 gallons of white liquor, a chemical used in paper processing, authorities said.
The Trump administration announced that people with temporary visas seeking to adjust their immigration status to obtain green cards must return to their home countries to “do so through consular processing,” in a major change from current practice. NBC News' Jonathan Allen explains the impact and how the new requirement will be enforced.
This sponsored article is brought to you by Wetour Robotics. A field technician on a wind turbine, harness clipped, both hands on a wrench, needs to send a command to the diagnostic device hanging at her belt. A logistics worker on a loading dock, gloves on, eyes on the pallet, needs to redirect a connected lift. A person using an assistive mobility device on a crowded street wants to nudge it forward without taking out a phone or speaking aloud. None of these moments call for a smarter robot. They call for a smarter way to be heard by the machines that already exist. The industry has been building from one side The past three years of Physical AI have been a story of remarkable progress on the robot side of the loop. Companies like Boston Dynamics, Figure, and Unitree have advanced actuators, locomotion, and dexterity to a level that would have seemed implausible a decade ago. Google DeepMind’s Gemini Robotics has redefined what vision-language-action models can do in unstructured settings. The trajectory of the hardware and the foundation models is real, and it is accelerating. But there is another side to this loop, and it has been treated as a solved problem for too long. The interface between humans and machines has defaulted, for 40 years, to three input modalities: screens, buttons, and voice. Each of those assumes the user can stop, look down, and translate intent into structured commands. That assumption breaks the moment the work moves into a real environment. On a turbine. On a dock. On a sidewalk. In any setting where hands are occupied, eyes are committed, or speaking is impractical, the conventional interface stack quietly fails. Spatial Intent Fusion is the simultaneous processing of three streams of human-centered information, namely spatial position, visual context, and gestural intent: Your body is the interface. The bottleneck on the human side of the loop is becoming as important as the one on the machine side. And solving it requires a different question. Not how do we make the robot more capable, but how do we let the human participate in the computing system as naturally as the robot already does. Wetour Robotics’ bet: put the human back into the computing loop Wetour Robotics is betting that the next architectural leap in Physical AI is not about making the robot more capable. It is about making the human a first-class node in the computing network, with the same kind of low-latency, high-fidelity participation that connected devices already enjoy. Wetour Robotics’ engineers frame the problem this way: a wristband that recognizes a gesture is not enough. A camera that recognizes a scene is not enough. The information a human carries about what they are about to do is distributed across multiple channels, including where their body is in space, what their eyes are attending to, and what their muscles are preparing to do, and any single channel observed in isolation is ambiguous. Reconstructing intent reliably means fusing those channels at the operating system level, with latency low enough that the loop feels closed rather than mediated. This approach has a name. Wetour Robotics calls it Spatial Intent Fusion: the simultaneous processing of three streams of human-centered information, namely spatial position, visual context, and gestural intent, fused into a single real-time command for any connected physical device. It is the technical implementation behind a simpler positioning statement the company uses externally: your body is the interface. Orchestra is a portable intelligent hub running the operating system that handles sensor fusion, intent inference, command translation, and safety arbitration. The reference compute platform is NVIDIA Jetson Orin Nano Super, which provides enough on-device inference capacity to keep the entire control loop at the edge, with no cloud dependency on the critical path. Wetour Robotics The architecture: three layers, four engines, one loop Orchestra is not a single device but a layered platform, designed from the start to be sensor-flexible and actuator-agnostic. The architecture decomposes into three perception layers and four coordination engines. Orchestra itself is the local compute and orchestration core: a portable intelligent hub running the operating system that handles sensor fusion, intent inference, command translation, and safety arbitration. The reference compute platform is NVIDIA Jetson Orin Nano Super, which provides enough on-device inference capacity to keep the entire control loop at the edge, with no cloud dependency on the critical path. Edge inference is non-negotiable for this application. Full-chain latency from biosignal acquisition to actuator command is held under 100 milliseconds, the envelope inside which closed-loop control feels natural rather than laggy. VisionLink handles visual and spatial perception. Cameras feed into vision models that identify objects, estimate distances, and track environmental context. VisionLink is designed not as a passive recognition layer but as a real-time command generator: its outputs feed directly into Orchestra OS to be fused with biosignal data. Conductor is the biosignal pipeline. It ingests raw surface electromyographic (sEMG) data from a wrist-worn device, classifies temporal patterns into discrete gestures or continuous control signals, and outputs actuator commands. The technically interesting property of sEMG for this use case is that the signal precedes visible motion. Motor unit action potentials appear at the skin surface roughly 50 to 80 milliseconds before a finger completes the corresponding gesture. Wetour Robotics calls this property pre-motion intent sensing, and it is what allows Orchestra to anticipate user intent rather than react to it. On top of the three perception layers, Orchestra OS runs four coordination engines. The Perception Engine ingests and normalizes raw sensor streams. The Intent Engine performs Spatial Intent Fusion across modalities, resolving what the user is trying to do given where they are, what they are looking at, and what their hand is signaling. The Orchestration Engine translates intent into device-specific command sequences for any connected actuator. The Safety Engine arbitrates conflicting commands, enforces operational envelopes, and gates execution against runtime safety conditions. Wetour Robotics The trade-offs we’re honest about No system that bridges the human body and the digital world is finished. Three engineering challenges remain open, and the company addresses each with a deliberate trade-off rather than a claim of having fully solved it. Baseline stability of sEMG under motion. In a stationary user, continuous gesture recognition from sEMG is reliable. Once the user is walking, climbing, or otherwise moving, motion artifacts and electrode drift degrade the signal in ways that are difficult to fully compensate for. Rather than overpromise on continuous control in dynamic settings, Orchestra defaults to a smaller set of robust discrete gestures in complex operating environments, and reserves continuous control modes for contexts where the signal-to-noise ratio supports them. Miniaturization of edge AI compute. Running the Orchestra control loop entirely at the edge requires real on-device inference, which has historically meant trading off between compute capacity, battery life, and form factor. Wetour Robotics’ approach has been a compact carrier board paired with a thermal design and a battery module sized for all-day wearability. The result is a hub that travels with the user rather than tethering them to a desk, and that performs the full perception-to-actuation loop without offloading to the cloud. Heterogeneity of third-party device protocols. The actuator side of the loop is a fragmented landscape. Different manufacturers expose different command interfaces, different communication stacks, and different safety conventions, and a Physical AI operating system has to integrate with all of them. Wetour Robotics uses an AI-agent layer to negotiate connection and protocol translation adaptively, so that Orchestra OS can ingest data from a wide range of devices, run them through neural network models that infer human intent, and emit the right command on the right protocol for the device on the other end. Why this matters, and why it helps the rest of the field The history of computing is a history of interface revolutions. Command lines gave way to graphical user interfaces, which gave way to touch, which gave way to voice. Each transition expanded who could participate in the system and what they could do with it. The next transition is not about a new screen or a new microphone. It is about treating the human body itself as a participant in the computing network, capable of contributing intent at the same speed and fidelity that any other connected node can. The history of computing is a history of interface revolutions. The next transition is not about a new screen or a new microphone — it is about treating the human body itself as a participant in the computing network. This path is not a competitor to the work being done on humanoid robots, foundation models for embodied AI, and dexterous manipulation. It is the missing complement to that work. The hardest open problem for humanoid systems is the data: every natural interaction between a human and the physical world is a potential training signal, and most of those interactions are currently invisible to any computing system. As more humans become first-class nodes in the loop, those interactions become observable, structured, and ultimately useful for training the next generation of embodied AI, including the humanoid robots being developed today. In other words: putting the human back into the computing loop is not just about better interfaces for individual users. It is about generating the kind of grounded, in-the-wild human-machine interaction data that the broader Physical AI ecosystem will need to keep advancing. The robot side and the human side of the loop are not two competing futures. They are two halves of the same one. That is what Wetour Robotics means when it says: Your body is the interface. Learn more at wetourrobotics.com.
A guide to ten technological components — from THz communications and AI/ML to reconfigurable intelligent surfaces — poised to define 6G wireless networks. What Attendees will Learn Which frequencies 6G will use — Understand why THz bands (above 100 GHz) and the7–24 GHz range are under consideration, what challenges CMOS technology faces at sub-THz frequencies, and how new semiconductor approaches aim to close the output-power gap for future link budgets. How AI/ML and joint communications and sensing reshape the air interface — how auto encoder-based end-to-end learning can replace traditional signal-processing blocks, and how a single waveform may serve both data transmission and radar-like environmental sensing. What reconfigurable intelligent surfaces and photonics bring to the radio environment— Explore how programmable metamaterial panels can steer and shape electromagnetic waves, and how visible light communications and all-photonics networks extend capacity and lower latency. How ultra-massive MIMO, full-duplex, and new network topologies enable a true 3D“network of networks” — Understand how antenna arrays with vastly more elements, simultaneously transmit/receive on the same frequency, and non-terrestrial nodes converge to deliver ubiquitous, high-capacity 6G coverage. Download this free whitepaper now!
This article is brought to you by DAIMON Robotics. This April, Hong Kong-based DAIMON Robotics has released Daimon-Infinity, which it describes as the largest omni-modal robotic dataset for physical AI, featuring high resolution tactile sensing and spanning a wide range of tasks from folding laundry at home to manufacturing on factory assembly lines. The project is supported by collaborative efforts of partners across China and the globe, including Google DeepMind, Northwestern University, and the National University of Singapore. The move signals a key strategic initiative for DAIMON, a two-and-a-half-year-old company known for its advanced tactile sensor hardware, most notably a monochromatic, vision-based tactile sensor that packs over 110,000 effective sensing units into a fingertip-sized module. Drawing on its high-resolution tactile sensing technology and a distributed out-of-lab collection network capable of generating millions of hours of data annually, DAIMON is building large-scale robot manipulation datasets that include vast amounts of tactile sensing data. To accelerate the real-world deployment of embodied AI, the company has also open-sourced 10,000 hours of its data. Prof. Michael Yu Wang, co-founder and chief scientist at DAIMON Robotics, has pioneered Vision-Tactile-Language-Action (VTLA) architecture, elevating the tactile to a modality on par with vision.DAIMON Robotics Behind the strategy is Prof. Michael Yu Wang, DAIMON’s co-founder and chief scientist. Prof. Wang earned his PhD at Carnegie Mellon — studying manipulation under Matt Mason — and went on to found the Robotics Institute at the Hong Kong University of Science and Technology. An IEEE Fellow and former Editor-in-Chief of IEEE Transactions on Automation Science and Engineering, he has spent roughly four decades in the field. His objective is to address the missing “insensitivity” of robot manipulation, which practically relies on the dominant Vision-Language-Action (VLA) model. He and his team have pioneered Vision-Tactile-Language-Action (VTLA) architecture, elevating the tactile to a modality on par with vision. We spoke with Prof. Wang about how tactile feedback aims to change dexterous manipulation, how the dataset initiative is foreseen to improve our understanding of robotic hands in natural environments, and where — from hotels to convenience stores in China — he sees touch-enabled robots making their first real-world inroads. Daimon-Infinity is the world’s largest omni-modal dataset for Physical AI, featuring million-hour scale multimodal data, ultra-high-res tactile feedback, data from 80+ real scenarios and 2,000+ human skills, and more.DAIMON Robotics The Dataset Initiative This month, DAIMON Robotics released the largest and most comprehensive robotic manipulation dataset with multiple leading academic institutions and enterprises. Why releasing the dataset now, rather than continuing to focus on product development? What impact will this have on the embodied intelligence industry? DAIMON Robotics has been around for almost two and a half years. We have been committed to developing high-resolution, multimodal tactile sensing devices to perceive the interaction between a robot’s hand (particularly its fingertips) and objects. Our devices have become quite robust. They are now accepted and used by a large segment of users, including academic and research institutes as well as leading humanoid robotics companies. As embodied AI continues to advance, the critical role of data has been clearer. Data scarcity remains a primary bottleneck in robot learning, particularly the lack of physical interaction data, which is essential for robots to operate effectively in the real world. Consequently, data quality, reliability, and cost have become major concerns in both research and commercial development. This is exactly where DAIMON excels. Our vision-based tactile technology captures high-quality, multimodal tactile data. Beyond basic contact forces, it records deformation, slip and friction, material properties and surface textures — enabling a comprehensive reconstruction of physical interactions. Building on our expertise in multimodal fusion, we have developed a robust data processing pipeline that seamlessly integrates tactile feedback with vision, motion trajectories, and natural language, transforming raw inputs into training-ready dataset for machine learning models. Recognizing the industry-wide data gap, we view large-scale data collection not only as our unique competitive advantage, but as a responsibility to the broader community. By building and open-sourcing the dataset, we aim to provide the high-quality “fuel” needed to power embodied AI, ultimately accelerating the real-world deployment of general-purpose robotic foundation models. The robotics industry is highly competitive, and many teams have chosen to focus on data. DAIMON is releasing a large and highly comprehensive cross-embodiment, vision-based tactile multimodal robotic manipulation dataset. How were you able to achieve this? We have a dedicated in-house team focused on expanding our capabilities, including building hardware devices and developing our own large-scale model. Although we are a relatively small company, our core tactile sensing technology and innovative data collection paradigm enable us to build large-scale dataset. Our approach is to broaden our offering. We have built the world’s largest distributed out-of-lab data collection network. Rather than relying on centralized data factories, this lightweight and scalable system allows data to be gathered across diverse real-world environments, enabling us to generate millions of hours of data per year. “To drive the advancement of the entire embodied AI field, we have open-sourced 10,000 hours of the dataset for the broader community.” —Prof. Michael Yu Wang, DAIMON Robotics This dataset is being jointly developed with several institutions worldwide. What roles did they play in its development, and how will the dataset benefit their research and products? Besides China based teams, our partners include leading research groups from universities, such as Northwestern University and the National University of Singapore, as well as top global enterprises like Google DeepMind and China Mobile. Their decision to partner with DAIMON is a strong testament to the value of our tactile-rich dataset. Among the companies involved there are some that have already built their own models but are now incorporating tactile information. By deploying our data collection devices across research, manufacturing and other real-world scenarios, they help us to gather highly practical, application-driven data. In turn, our partners leverage the data to train models tailored to their specific use cases. Furthermore, to drive the advancement of the entire embodied AI field, we have open-sourced 10,000 hours of the dataset for the broader community. Equipped with Daimon’s visuotactile sensor, the gripper delicately senses contact and precisely controls force to pick up a fragile eggshell.Daimon Robotics From VLA to VTLA: Why Tactile Sensing Changes the Equation The mainstream paradigm in robotics is currently the Vision-Language-Action (VLA) model, but your team has proposed a Vision-Tactile-Language-Action (VTLA) model. Why is it necessary to incorporate tactile sensing? What does it enable robots to achieve, and which tasks are likely to fail without tactile feedback? Over these years of working to make generalist robots capable of performing manipulation tasks, especially dexterous manipulation — not just power grasping or holding an object, but manipulating objects and using tools to impart forces and motion onto parts — we see these robots being used in household as well as industrial assembly settings. It is well established that tactile information is essential for providing feedback about contact states so that robots can guide their hands and fingers to perform reliable manipulation. Without tactile sensing, robots are severely limited. They struggle to locate objects in dark environments, and without slip detection, they can easily drop fragile items like glass. Furthermore, the inability to precisely control force often leads to failed manipulation tasks or, in severe cases, physical damage. Naturally, the VLA approach needs to be enhanced to incorporate tactile information. We expanded the VLA framework to incorporate tactile data, creating the VTLA model. An additional benefit of our tactile sensor is that it is vision-based: We capture visual images of the deformation on the fingertip surface. We capture multiple images in a time sequence that encodes contact information, from which we can infer forces and other contact states. This aligns well with the visual framework that VLA is based upon. Having tactile information in a visual image format makes it naturally suitable for integration into the VLA framework, transforming it into a VTLA system. That is the key advantage: Vision-based tactile sensors provide very high resolution at the pixel level, and this data can be incorporated into the framework, whether it is an end-to-end model or another type of architecture. DAIMON has been known for its vision-based tactile sensors that can pack over 110,000 effective sensing units.DAIMON Robotics The Technology: Monochromatic Vision-based Tactile Sensing You and your team have spent many years deeply engaged in vision-based tactile sensing and have developed the world’s first monochromatic vision-based tactile sensing technology. Why did you choose this technical path? Once we started investigating tactile sensors, we understood our needs. We wanted sensors that closely mimic what we have under our fingertip skin. Physiological studies have well documented the capabilities humans have at their fingertips — knowing what we touch, what kind of material it is, how forces are distributed, and whether it is moving into the right position as our brain controls our hands. We knew that replicating these capabilities on a robot hand’s fingertips would help considerably. When we surveyed existing technologies, we found many types, including vision-based tactile sensors with tri-color optics and other simpler designs. We decided to integrate the best of these into an engineering-robust solution that works well without being overly complicated, keeping cost, reliability, and sensitivity within a satisfactory range, thus ultimately developing a monochromatic vision-based tactile sensing technique. This is fundamentally an engineering approach rather than a purely scientific one, since a great deal of foundational research already existed. With the growing realization of the necessity of tactile data, all of this will advance hand in hand. DAIMON vision-based tactile sensor captures high-quality, multimodal tactile data.DAIMON Robotics Last year, DAIMON launched a multi-dimensional, high-resolution, high-frequency vision-based tactile sensor. Compared with traditional tactile sensors, where does its core advantage lie? Which industries could it potentially transform? The key features of our sensors are the density of distributed force measurement and the deformation we can capture over the area of a fingertip. I believe we have the highest density in terms of sensing units. That is one very important metric. The other is dynamics: the frequency and bandwidth — how quickly we can detect force changes, transmit signals, and process them in real time. Other important aspects are largely engineering-related, such as reliability, drift, durability of the soft surface, and resistance to interference from magnetic, optical, or environmental factors. A growing number of researchers and companies are recognizing the importance of tactile sensing and adopting our technology. I believe the advances in tactile sensing will elevate the entire community and industry to a higher level. One of our potential customers is deploying humanoid robots in a small convenience store, with densely packed shelves where shelf space is at a premium. The robot needs to reach into very tight spaces — tighter than books on a shelf — to pick out an object. Current two-jaw parallel grippers cannot fit into most of these spaces. Observing how humans pick up objects, you clearly need at least three slim fingers to touch and roll the object toward you and secure it. Thus, we are starting to see very specific needs where tactile sensing capabilities are essential. From Academia to Startup After 40 years in academia — founding the HKUST Robotics Institute, earning prestigious honors including IEEE Fellow, and serving as Editor-in-Chief of IEEE TASE — what motivated you to found DAIMON Robotics? I have come a long way. I started learning robotics during my PhD at Carnegie Mellon, where there were truly remarkable groups working on locomotion under Marc Raibert, who founded Boston Dynamics, and on manipulation under my advisor, Matt Mason, a leader in the field. We have been working on dexterous manipulation, not only at Carnegie Mellon, but globally for many years. However, progress has been limited for a long time, especially in building dexterous hands and making them work. Only recently have locomotion robots truly taken off, and only in the last few years have we begun to see major advancements in robot hands. There is clearly room for advancing manipulation capabilities, which would enable robots to do work like humans. While at Hong Kong University of Science and Technology, I saw increasingly greater people entering this area in the form of students and postdoctoral researchers. We wanted to jumpstart our effort by leveraging the available capital and talent resources. Fortunately, one of my postdocs, Dr. Duan Jianghua, has a strong sense for commercial opportunities. Recognizing the rapid growth of robotics market and the unique value that our vision-based tactile sensing technology could bring, together we started DAIMON Robotics, and it has progressed well. The community has grown tremendously in China, Japan, Korea, the U.S., and Europe. Robots equipped with DAIMON technology have been deployed in factory settings. The company aims to enable robots to achieve “embodied intelligence” and close the gap between what they can see and what they can feel.DAIMON Robotics Business Model and Commercial Strategy What is DAIMON’s current business model and strategic focus? What role does the dataset release play in your commercial strategy? We started as a device company focused on making highly capable tactile sensors, especially for robot hands. But as technology and business developed, everyone realized it is not just about one component, rather the entire technology chain: devices, data of adequate quality and quantity, and finally the right framework to build, train, and deploy models on robots in real application environments. Our business strategy is best described as “3D”: Devices, Data, and Deployment. We build devices for data collection, our own ecosystem, and for deploying them in our partners’ potential application domains. This enables the collection of real-world tactile-rich data and complete closed-loop validation. This will become an integral part of the 3D business model. Most startups in this space are following a similar path until eventually some may become more specialized or more tightly integrated with other companies. For now, it is mostly vertical integration. Embodied Skills and the Convergence Moment You’ve introduced the concept of “embodied skills” as essential for humanoid robots to move beyond having just an advanced AI “brain.” What prompted this insight? What new capabilities could embodied skills enable? After the rapid evolution of models and hardware over the past two years, has your definition or roadmap for embodied skills evolved? We have come a long way now see a convergence point where electrical, electronic, and mechatronic hardware technologies have advanced tremendously in last two decades. Robots are now fully electric, do not require hydraulics, because hardware has evolved rapidly. Modern electronics provide tremendous bandwidth with high torques. If we can build intelligence into these systems, we can create truly humanoid robots with the ability to operate in unstructured environments, make decisions, and take actions autonomously. “Our vision is for robots to achieve robust manipulation capabilities and evolve into reliable partners for humans.” —Prof. Michael Yu Wang, DAIMON Robotics AI has arrived at exactly the right time. Enormous resources have been invested in AI development, especially large language models, which are now being generalized into world models that enable physical AI capabilities. We would like to see these manifested in real-world systems. While both AI and core hardware technologies continue to evolve, the focus is much clearer now. For example, human-sized robots are preferred in a home environment. This is an exciting domain with a promise of great societal benefit if we can eventually achieve safe, reliable, and cost-effective robots. The Road to Real-World Deployment Today, many robots can deliver impressive demos, yet there remains a gap before they truly enter real-world applications. What could be a potential trigger for real-world deployment? Which scenarios are most likely to achieve large-scale deployment first? I think the road toward large-scale deployment of generalist robots is still long, but we are starting to see signs of feasibility within specific domains. It is very similar to autonomous vehicles, where we are yet to see full deployment of robo-taxis, while we have already started to find mobile robots and smaller vehicles widely deployed in the hospitality industry. Virtually every major hotel in China now has a delivery robot — no arms, just a vehicle that picks up items from the hotel lobby (e.g., food deliveries). The delivery person just loads the food and selects the room number. It is up to the robot thereafter to navigate and reach the guest’s room, which includes using the elevator, to deliver the food. This is already nearly 100 percent deployed in major Chinese hotels. Hotel and restaurant robots are viewed as a model for deploying humanoid robots in specific domains like overnight drugstores and convenience stores. I expect complete deployment in such settings within a short timeframe, followed by other applications. Overall, we can expect autonomous robots, including humanoids, to progressively penetrate specific sectors, delivering value in each and expanding into others. Ultimately, our vision is for robots to achieve robust manipulation capabilities and evolve into reliable partners for humans. By seamlessly integrating into our homes and daily lives, they will genuinely benefit and serve humanity. This interview has been edited for length and clarity.
When it comes to AI models, size matters. Even though some artificial-intelligence experts warn that scaling up large language models (LLMs) is hitting diminishing performance returns, companies are still coming out with ever larger AI tools. Meta’s latest Llama release had a staggering 2 trillion parameters that define the model. As models grow in size, their capabilities increase. But so do the energy demands and the time it takes to run the models, which increases their carbon footprint. To mitigate these issues, people have turned to smaller, less capable models and using lower-precision numbers whenever possible for the model parameters. But there is another path that may retain a staggeringly large model’s high performance while reducing the time it takes to run an energy footprint. This approach involves befriending the zeros inside large AI models. For many models, most of the parameters—the weights and activations—are actually zero, or so close to zero that they could be treated as such without losing accuracy. This quality is known as sparsity. Sparsity offers a significant opportunity for computational savings: Instead of wasting time and energy adding or multiplying zeros, these calculations could simply be skipped; rather than storing lots of zeros in memory, one need only store the nonzero parameters. Unfortunately, today’s popular hardware, like multicore CPUs and GPUs, do not naturally take full advantage of sparsity. To fully leverage sparsity, researchers and engineers need to rethink and re-architect each piece of the design stack, including the hardware, low-level firmware, and application software. In our research group at Stanford University, we have developed the first (to our knowledge) piece of hardware that’s capable of calculating all kinds of sparse and traditional workloads efficiently. The energy savings varied widely over the workloads, but on average our chip consumed one-seventieth the energy of a CPU, and performed the computation on average eight times as fast. To do this, we had to engineer the hardware, low-level firmware, and software from the ground up to take advantage of sparsity. We hope this is just the beginning of hardware and model development that will allow for more energy-efficient AI. What is sparsity? Neural networks, and the data that feeds into them, are represented as arrays of numbers. These arrays can be one-dimensional (vectors), two-dimensional (matrices), or more (tensors). A sparse vector, matrix, or tensor has mostly zero elements. The level of sparsity varies, but when zeroes make up more than 50 percent of any type of array, it can stand to benefit from sparsity-specific computational methods. In contrast, an object that is not sparse—that is, it has few zeros compared with the total number of elements—is called dense. Sparsity can be naturally present, or it can be induced. For example, a social-network graph will be naturally sparse. Imagine a graph where each node (point) represents a person, and each edge (a line segment connecting the points) represents a friendship. Since most people are not friends with one another, a matrix representing all possible edges will be mostly zeros. Other popular applications of AI, such as other forms of graph learning and recommendation models, contain naturally occurring sparsity as well. Beyond naturally occurring sparsity, sparsity can also be induced within an AI model in several ways. Two years ago, a team at Cerebras showed that one can set up to 70 to 80 percent of parameters in an LLM to zero without losing any accuracy. Cerebras demonstrated these results specifically on Meta’s open-source Llama 7B model, but the ideas extend to other LLM models like ChatGPT and Claude. The case for sparsity Sparse computation’s efficiency stems from two fundamental properties: the ability to compress away zeros and the convenient mathematical properties of zeros. Both the algorithms used in sparse computation and the hardware dedicated to them leverage these two basic ideas. First, sparse data can be compressed, making it more memory efficient to store “sparsely”—that is, in something called a sparse data type. Compression also makes it more energy efficient to move data when dealing with large amounts of it. This is best understood by an example. Take a four-by-four matrix with three nonzero elements. Traditionally, this matrix would be stored in memory as is, taking up 16 spaces. This matrix can also be compressed into a sparse data type, getting rid of the zeros and saving only the nonzero elements. In our example, this results in 13 memory spaces as opposed to 16 for the dense, uncompressed version. These savings in memory increase with increased sparsity and matrix size. In addition to the actual data values, compressed data also requires metadata. The row and column locations of the nonzero elements also must be stored. This is usually thought of as a “fibertree”: The row labels containing nonzero elements are listed and linked to the column labels of the nonzero elements, which are then linked to the values stored in those elements. In memory, things get a bit more complicated still: The row and column labels for each nonzero value must be stored as well as the “segments” that indicate how many such labels to expect, so the metadata and data can be clearly delineated from one another. In a dense, noncompressed matrix data type, values can be accessed either one at a time or in parallel, and their locations can be calculated directly with a simple equation. However, accessing values in sparse, compressed data requires looking up the coordinates of the row index and using that information to “indirectly” look up the coordinates of the column index before finally reaching the value. Depending on the actual locations of the sparse data values, these indirect lookups can be extremely random, making the computation data-dependent and requiring the allocation of memory lookups on the fly. Second, two mathematical properties of zero let software and hardware skip a lot of computation. Multiplying any number by zero will result in a zero, so there’s no need to actually do the multiplication. Adding zero to any number will always return that number, so there’s no need to do the addition either. In matrix-vector multiplication, one of the most common operations in AI workloads, all computations except those involving two nonzero elements can simply be skipped. Take, for example, the four-by-four matrix from the previous example and a vector of four numbers. In dense computation, each element of the vector must be multiplied by the corresponding element in each row and then added together to compute the final vector. In this case, that would take 16 multiplication operations and 16 additions (or four accumulations). In sparse computation, only the nonzero elements of the vector need be considered. For each nonzero vector element, indirect lookup can be used to find any corresponding nonzero matrix element, and only those need to be multiplied and added. In the example shown here, only two multiplication steps will be performed, instead of 16. The trouble with GPUs and CPUs Unfortunately, modern hardware is not well suited to accelerating sparse computation. For example, say we want to perform a matrix-vector multiplication. In the simplest case, in a single CPU core, each element in the vector would be multiplied sequentially and then written to memory. This is slow, because we can do only one multiplication at a time. So instead people use CPUs with vector support or GPUs. With this hardware, all elements would be multiplied in parallel, greatly speeding up the application. Now, imagine that both the matrix and vector contain extremely sparse data. The vectorized CPU and GPU would spend most of their efforts multiplying by zero, performing completely ineffectual computations. Newer generations of GPUs are capable of taking some advantage of sparsity in their hardware, but only a particular kind, called structured sparsity. Structured sparsity assumes that two out of every four adjacent parameters are zero. However, some models benefit more from unstructured sparsity—the ability for any parameter (weight or activation) to be zero and compressed away, regardless of where it is and what it is adjacent to. GPUs can run unstructured sparse computation in software, for example, through the use of the cuSparse GPU library. However, the support for sparse computations is often limited, and the GPU hardware gets underutilized, wasting energy-intensive computations on overhead. Petra Péterffy When doing sparse computations in software, modern CPUs may be a better alternative to GPU computation, because they are designed to be more flexible. Yet, sparse computations on the CPU are often bottlenecked by the indirect lookups used to find nonzero data. CPUs are designed to “prefetch” data based on what they expect they’ll need from memory, but for randomly sparse data, that process often fails to pull in the right stuff from memory. When that happens, the CPU must waste cycles calling for the right data. Apple was the first to speed up these indirect lookups by supporting a method called an array-of-pointers access pattern in the prefetcher of their A14 and M1 chips. Although innovations in prefetching make Apple CPUs more competitive for sparse computation, CPU architectures still have fundamental overheads that a dedicated sparse computing architecture would not, because they need to handle general-purpose computation. Other companies have been developing hardware that accelerates sparse machine learning as well. These include Cerebras’s Wafer Scale Engine and Meta’s Training and Inference Accelerator (MTIA). The Wafer Scale Engine, and its corresponding sparse programming framework, have shown incredibly sparse results of up to 70 percent sparsity on LLMs. However, the company’s hardware and software solutions support only weight sparsity, not activation sparsity, which is important for many applications. The second version of the MTIA claims a sevenfold sparse compute performance boost over the MTIA v1. However, the only publicly available information regarding sparsity support in the MTIA v2 is for matrix multiplication, not for vectors or tensors. Although matrix multiplications take up the majority of computation time in most modern ML models, it’s important to have sparsity support for other parts of the process. To avoid switching back and forth between sparse and dense data types, all of the operations should be sparse. Onyx Instead of these halfway solutions, our team at Stanford has developed a hardware accelerator, Onyx, that can take advantage of sparsity from the ground up, whether it’s structured or unstructured. Onyx is the first programmable accelerator to support both sparse and dense computation; it’s capable of accelerating key operations in both domains. To understand Onyx, it is useful to know what a coarse-grained reconfigurable array (CGRA) is and how it compares with more familiar hardware, like CPUs and field-programmable gate arrays (FPGAs). CPUs, CGRAs, and FPGAs represent a trade-off between efficiency and flexibility. Each individual logic unit of a CPU is designed for a specific function that it performs efficiently. On the other hand, since each individual bit of an FPGA is configurable, these arrays are extremely flexible, but very inefficient. The goal of CGRAs is to achieve the flexibility of FPGAs with the efficiency of CPUs. CGRAs are composed of efficient and configurable units, typically memory and compute, that are specialized for a particular application domain. This is the key benefit of this type of array: Programmers can reconfigure the internals of a CGRA at a high level, making it more efficient than an FPGA but more flexible than a CPU. The Onyx chip, built on a coarse-grained reconfigurable array (CGRA), is the first (to our knowledge) to support both sparse and dense computations. Olivia Hsu Onyx is composed of flexible, programmable processing element (PE) tiles and memory (MEM) tiles. The memory tiles store compressed matrices and other data formats. The processing element tiles operate on compressed matrices, eliminating all unnecessary and ineffectual computation. The Onyx compiler handles conversion from software instructions to CGRA configuration. First, the input expression—for instance, a sparse vector multiplication—is translated into a graph of abstract memory and compute nodes. In this example, there are memories for the input vectors and output vectors, a compute node for finding the intersection between nonzero elements, and a compute node for the multiplication. The compiler figures out how to map the abstract memory and compute nodes onto MEMs and PEs on the CGRA, and then how to route them together so that they can transfer data between them. Finally, the compiler produces the instruction set needed to configure the CGRA for the desired purpose. Since Onyx is programmable, engineers can map many different operations, such as vector-vector element multiplication, or the key tasks in AI, like matrix-vector or matrix-matrix multiplication, onto the accelerator. We evaluated the efficiency gains of our hardware by looking at the product of energy used and the time it took to compute, called the energy-delay product (EDP). This metric captures the trade-off of speed and energy. Minimizing just energy would lead to very slow devices, and minimizing speed would lead to high-area, high-power devices. Onyx achieves up to 565 times as much energy-delay product over CPUs (we used a 12-core Intel Xeon CPU) that utilize dedicated sparse libraries. Onyx can also be configured to accelerate regular, dense applications, similar to the way a GPU or TPU would. If the computation is sparse, Onyx is configured to use sparse primitives, and if the computation is dense, Onyx is reconfigured to take advantage of parallelism, similar to how GPUs function. This architecture is a step toward a single system that can accelerate both sparse and dense computations on the same silicon. Just as important, Onyx enables new algorithmic thinking. Sparse acceleration hardware will not only make AI more performance- and energy efficient but also enable researchers and engineers to explore new algorithms that have the potential to dramatically improve AI. The future with sparsity Our team is already working on next-generation chips built off of Onyx. Beyond matrix multiplication operations, machine learning models perform other types of math, like nonlinear layers, normalization, the softmax function, and more. We are adding support for the full range of computations on our next-gen accelerator and within the compiler. Since sparse machine learning models may have both sparse and dense layers, we are also working on integrating the dense and sparse accelerator architecture more efficiently on the chip, allowing for fast transformation between the different data types. We’re also looking at ways to manage memory constraints by breaking up the sparse data more effectively so we can run computations on several sparse accelerator chips. We are also working on systems that can predict the performance of accelerators such as ours, which will help in designing better hardware for sparse AI. Longer term, we’re interested in seeing whether high degrees of sparsity throughout AI computation will catch on with more model types, and whether sparse accelerators become adopted at a larger scale. Building the hardware to unstructured sparsity and optimally take advantage of zeros is just the beginning. With this hardware in hand, AI researchers and engineers will have the opportunity to explore new models and algorithms that leverage sparsity in novel and creative ways. We see this as a crucial research area for managing the ever-increasing runtime, costs, and environmental impact of AI.
Many of the world’s most advanced electronic systems—including Internet routers, wireless base stations, medical imaging scanners, and some artificial intelligence tools—depend on field-programmable gate arrays. Computer chips with internal hardware circuits, the FPGAs can be reconfigured after manufacturing. On 12 March, an IEEE Milestone plaque recognizing the first FPGA was dedicated at the Advanced Micro Devices campus in San Jose, Calif., the former Xilinx headquarters and the birthplace of the technology. The FPGA earned the Milestone designation because it introduced iteration to semiconductor design. Engineers could redesign hardware repeatedly without fabricating a new chip, dramatically reducing development risk and enabling faster innovation at a time when semiconductor costs were rising rapidly. The ceremony, which was organized by the IEEE Santa Clara Valley Section, brought together professionals from across the semiconductor industry and IEEE leadership. Speakers at the event included Stephen Trimberger, an IEEE and ACM Fellow whose technical contributions helped shape modern FPGA architecture. Trimberger reflected on how the invention enabled software-programmable hardware. Solving computing’s flexibility-performance tradeoff FPGAs emerged in the 1980s to address a core limitation in computing. A microprocessor executes software instructions sequentially, making it flexible but sometimes too slow for workloads requiring many operations at once. At the other extreme, application-specific integrated circuits are chips designed to do only one task. ASICs achieve high efficiency but require lengthy development cycles and nonrecurring engineering costs, which are large, upfront investments. Expenses include designing the chip and preparing it for manufacturing—a process that involves creating detailed layouts, building masks for the fabrication machines, and setting up production lines to handle the tiny circuits. “ASICs can deliver the best performance, but the development cycle is long and the nonrecurring engineering cost can be very high,” says Jason Cong, an IEEE Fellow and professor of computer science at the University of California, Los Angeles. “FPGAs provide a sweet spot between processors and custom silicon.” Cong’s foundational work in FPGA design automation and high-level synthesis transformed how reconfigurable systems are programmed. He developed synthesis tools that translate C/C++ into hardware designs, for example. At the heart of his work is an underlying principle first espoused by electrical engineer Ross Freeman: By configuring hardware using programmable memory embedded inside the chip, FPGAs combine hardware-level speed with the adaptability traditionally associated with software. Silicon Valley origins: the first FPGA The FPGA architecture originated in the mid-1980s at Xilinx, a Silicon Valley company founded in 1984. The invention is widely credited to Freeman, a Xilinx cofounder and the startup’s CTO. He envisioned a chip with circuitry that could be configured after fabrication rather than fixed permanently during creation. Articles about the history of the FPGA emphasize that he saw it as a deliberate break from conventional chip design. At the time, semiconductor engineers treated transistors as scarce resources. Custom chips were carefully optimized so that nearly every transistor served a specific purpose. Freeman proposed a different approach. He figured Moore’s Law would soon change chip economics. The principle holds that transistor counts roughly double every two years, making computing cheaper and more powerful. Freeman posited that as transistors became abundant, flexibility would matter more than perfect efficiency. He envisioned a device composed of programmable logic blocks connected through configurable routing—a chip filled with what he described as “open gates,” ready to be defined by users after manufacturing. Instead of fixing hardware in silicon permanently, engineers could configure and reconfigure circuits as requirements evolved. Freeman sometimes compared the concept to a blank cassette tape: Manufacturers would supply the medium, while engineers determined its function. The analogy captured a profound shift in who controls the technology, shifting hardware design flexibility from chip fabrication facilities to the system designers themselves. In 1985 Xilinx introduced the first FPGA for commercial sale: the XC2064. The device contained 64 configurable logic blocks—small digital circuits capable of performing logical operations—arranged in an 8-by-8 grid. Programmable routing channels allowed engineers to define how signals moved between blocks, effectively wiring a custom circuit with software. Fabricated using a 2-micrometer process (meaning that 2 µm was the minimum size of the features that could be patterned onto silicon using photolithography), the XC2064 implemented a few thousand logic gates. Modern FPGAs can contain hundreds of millions of gates, enabling vastly more complex designs. Yet the XC2064 established a design workflow still used today: Engineers describe the hardware behavior digitally and then “compile the design,” a process that automatically translates the plans into the instructions the FPGA needs to set its logic blocks and wiring, according to AMD. Engineers then load that configuration onto the chip. The breakthrough: hardware defined by memory Earlier programmable logic devices, such as erasable programmable read-only memory, or EPROM, allowed limited customization but relied on largely fixed wiring structures that did not scale well as circuits grew more complex, Cong says. FPGAs introduced programmable interconnects—networks of electronic switches controlled by memory cells distributed across the chip. When powered on, the device loads a bitstream configuration file that determines how its internal circuits behave. “As process technology improved and transistor counts increased, the cost of programmability became much less significant,” Cong says. From “glue logic” to essential infrastructure “Initially, FPGAs were used as what engineers called glue logic,” Cong says. Glue logic refers to simple circuits that connect processors, memory, and peripheral devices so the system works reliably, according to PC Magazine. In other words, it “glues” different components together, especially when interfaces change frequently. Early adopters recognized the advantage of hardware that could adapt as standards evolved. In “The History, Status, and Future of FPGAs,” published in Communications of the ACM, engineers at Xilinx and organizations such as Bell Labs, Fairchild Semiconductor, IBM, and Sun Microsystems said the earliest uses of FPGAs were for prototyping ASICs. They also used it for validating complex systems by running their software before fabrication, allowing the companies to deploy specialized products manufactured in modest volumes. Those uses revealed a broader shift: Hardware no longer needed to remain fixed once deployed. Attendees at the Milestone plaque dedication ceremony included (seated L to R) 2025 IEEE President Kathleen Kramer, 2024 IEEE President Tom Coughlin, and Santa Clara Valley Section Milestones Chair Brian Berg.Douglas Peck/AMD Semiconductor economics changed the equation The rise of FPGAs closely followed changes in semiconductor economics, Cong says. Developing a custom chip requires a large upfront investment before production begins. As fabrication costs increased, products had to ship in large quantities to make ASIC development economically viable, according to a post published by AnySilicon. FPGAs allowed designers to move forward without that larger monetary commitment. ASIC development typically requires 18 to 24 months from conception to silicon, while FPGA implementations often can be completed within three to six months using modern design tools, Cong says. The shorter cycle and the ability to reconfigure the hardware enabled startups, universities, and equipment manufacturers to experiment with advanced architectures that were previously accessible mainly to large chip companies. Lookup tables and the rise of reconfigurable computing A popular technique for implementing mathematical functions in hardware is the lookup table (LUT). A LUT is a small memory element that stores the results of logical operations, according to “LUT-LLM: Efficient Large Language Model Inference with Memory-based Computations on FPGAs,” a paper selected for presentation next month at the 34th IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM). Instead of repeatedly recalculating outcomes, the chip retrieves answers directly from memory. Cong compares the approach to consulting multiplication tables rather than recomputing the arithmetic each time. Research led by Cong and others helped develop efficient methods for mapping digital circuits onto LUT-based architectures, shaping routing and layout strategies used in modern devices. As transistor budgets expanded, FPGA vendors integrated memory blocks, digital signal-processing units, high-speed communication interfaces, cryptographic engines, and embedded processors, transforming the devices into versatile computing platforms. Why the gate arrays are distinct from CPUs, GPUs, and ASICs FPGAs coexist with other processors because each one optimizes different priorities. Central processing units excel at general computing. Graphics processing units, designed to perform many calculations simultaneously, dominate large parallel workloads such as AI training. ASICs provide maximum efficiency when designs remain stable and production volumes are high. “ASICs can deliver the best performance, but the development cycle is long, and the nonrecurring engineering cost can be very high. FPGAs provide a sweet spot between processors and custom silicon.” —Jason Cong, IEEE Fellow and professor of computer science at UCLA. “FPGAs are not replacements for CPUs or GPUs,” Cong says. “They complement those processors in heterogeneous computing systems.” Modern computing platforms increasingly combine multiple types of processors to balance flexibility, performance, and energy efficiency. A Milestone for an idea, not just a device This IEEE Milestone recognizes more than a successful semiconductor product. It also acknowledges a shift in how engineers innovate. Reconfigurable hardware allows designers to test ideas quickly, refine architectures, and deploy systems while standards and markets evolve. “Without FPGAs,” Cong says, “the pace of hardware innovation would likely be much slower.” Four decades after the first FPGA appeared, the technology’s enduring legacy reflects Freeman’s insight: Hardware did not need to remain fixed. By accepting a small amount of unused silicon in exchange for adaptability, engineers transformed chips from static products into platforms for continuous experimentation—turning silicon itself into a medium engineers could rewrite. Among those who attended the Milestone ceremony were 2025 IEEE President Kathleen Kramer; 2024 IEEE President Tom Coughlin; Avery Lu, chair of the IEEE Santa Clara Valley Section; and Brian Berg, history and milestones chair of IEEE Region 6. They joined AMD’s chief executive, Lisa Su, and Salil Raje, senior vice president and general manager of adaptive and embedded computing at AMD. The IEEE Milestone plaque honoring the field-programmable gate array reads: “The FPGA is an integrated circuit with user-programmable Boolean logic functions and interconnects. FPGA inventor Ross Freeman cofounded Xilinx to productize his 1984 invention, and in 1985 the XC2064 was introduced with 64 programmable 4-input logic functions. Xilinx’s FPGAs helped accelerate a dramatic industry shift wherein ‘fabless’ companies could use software tools to design hardware while engaging ‘foundry’ companies to handle the capital-intensive task of manufacturing the software-defined hardware.” Administered by the IEEE History Center and supported by donors, the IEEE Milestone program recognizes outstanding technical developments worldwide that are at least 25 years old. Check out Spectrum’s History of Technology channel to read more stories about key engineering achievements.
Andrew Ng has serious street cred in artificial intelligence. He pioneered the use of graphics processing units (GPUs) to train deep learning models in the late 2000s with his students at Stanford University, cofounded Google Brain in 2011, and then served for three years as chief scientist for Baidu, where he helped build the Chinese tech giant’s AI group. So when he says he has identified the next big shift in artificial intelligence, people listen. And that’s what he told IEEE Spectrum in an exclusive Q&A. Ng’s current efforts are focused on his company Landing AI, which built a platform called LandingLens to help manufacturers improve visual inspection with computer vision. He has also become something of an evangelist for what he calls the data-centric AI movement, which he says can yield “small data” solutions to big issues in AI, including model efficiency, accuracy, and bias. Andrew Ng on... What’s next for really big models The career advice he didn’t listen to Defining the data-centric AI movement Synthetic data Why Landing AI asks its customers to do the work The great advances in deep learning over the past decade or so have been powered by ever-bigger models crunching ever-bigger amounts of data. Some people argue that that’s an unsustainable trajectory. Do you agree that it can’t go on that way? Andrew Ng: This is a big question. We’ve seen foundation models in NLP [natural language processing]. I’m excited about NLP models getting even bigger, and also about the potential of building foundation models in computer vision. I think there’s lots of signal to still be exploited in video: We have not been able to build foundation models yet for video because of compute bandwidth and the cost of processing video, as opposed to tokenized text. So I think that this engine of scaling up deep learning algorithms, which has been running for something like 15 years now, still has steam in it. Having said that, it only applies to certain problems, and there’s a set of other problems that need small data solutions. When you say you want a foundation model for computer vision, what do you mean by that? Ng: This is a term coined by Percy Liang and some of my friends at Stanford to refer to very large models, trained on very large data sets, that can be tuned for specific applications. For example, GPT-3 is an example of a foundation model [for NLP]. Foundation models offer a lot of promise as a new paradigm in developing machine learning applications, but also challenges in terms of making sure that they’re reasonably fair and free from bias, especially if many of us will be building on top of them. What needs to happen for someone to build a foundation model for video? Ng: I think there is a scalability problem. The compute power needed to process the large volume of images for video is significant, and I think that’s why foundation models have arisen first in NLP. Many researchers are working on this, and I think we’re seeing early signs of such models being developed in computer vision. But I’m confident that if a semiconductor maker gave us 10 times more processor power, we could easily find 10 times more video to build such models for vision. Having said that, a lot of what’s happened over the past decade is that deep learning has happened in consumer-facing companies that have large user bases, sometimes billions of users, and therefore very large data sets. While that paradigm of machine learning has driven a lot of economic value in consumer software, I find that that recipe of scale doesn’t work for other industries. Back to top It’s funny to hear you say that, because your early work was at a consumer-facing company with millions of users. Ng: Over a decade ago, when I proposed starting the Google Brain project to use Google’s compute infrastructure to build very large neural networks, it was a controversial step. One very senior person pulled me aside and warned me that starting Google Brain would be bad for my career. I think he felt that the action couldn’t just be in scaling up, and that I should instead focus on architecture innovation. “In many industries where giant data sets simply don’t exist, I think the focus has to shift from big data to good data. Having 50 thoughtfully engineered examples can be sufficient to explain to the neural network what you want it to learn.” —Andrew Ng, CEO & Founder, Landing AI I remember when my students and I published the first NeurIPS workshop paper advocating using CUDA, a platform for processing on GPUs, for deep learning—a different senior person in AI sat me down and said, “CUDA is really complicated to program. As a programming paradigm, this seems like too much work.” I did manage to convince him; the other person I did not convince. I expect they’re both convinced now. Ng: I think so, yes. Over the past year as I’ve been speaking to people about the data-centric AI movement, I’ve been getting flashbacks to when I was speaking to people about deep learning and scalability 10 or 15 years ago. In the past year, I’ve been getting the same mix of “there’s nothing new here” and “this seems like the wrong direction.” Back to top How do you define data-centric AI, and why do you consider it a movement? Ng: Data-centric AI is the discipline of systematically engineering the data needed to successfully build an AI system. For an AI system, you have to implement some algorithm, say a neural network, in code and then train it on your data set. The dominant paradigm over the last decade was to download the data set while you focus on improving the code. Thanks to that paradigm, over the last decade deep learning networks have improved significantly, to the point where for a lot of applications the code—the neural network architecture—is basically a solved problem. So for many practical applications, it’s now more productive to hold the neural network architecture fixed, and instead find ways to improve the data. When I started speaking about this, there were many practitioners who, completely appropriately, raised their hands and said, “Yes, we’ve been doing this for 20 years.” This is the time to take the things that some individuals have been doing intuitively and make it a systematic engineering discipline. The data-centric AI movement is much bigger than one company or group of researchers. My collaborators and I organized a data-centric AI workshop at NeurIPS, and I was really delighted at the number of authors and presenters that showed up. You often talk about companies or institutions that have only a small amount of data to work with. How can data-centric AI help them? Ng: You hear a lot about vision systems built with millions of images—I once built a face recognition system using 350 million images. Architectures built for hundreds of millions of images don’t work with only 50 images. But it turns out, if you have 50 really good examples, you can build something valuable, like a defect-inspection system. In many industries where giant data sets simply don’t exist, I think the focus has to shift from big data to good data. Having 50 thoughtfully engineered examples can be sufficient to explain to the neural network what you want it to learn. When you talk about training a model with just 50 images, does that really mean you’re taking an existing model that was trained on a very large data set and fine-tuning it? Or do you mean a brand new model that’s designed to learn only from that small data set? Ng: Let me describe what Landing AI does. When doing visual inspection for manufacturers, we often use our own flavor of RetinaNet. It is a pretrained model. Having said that, the pretraining is a small piece of the puzzle. What’s a bigger piece of the puzzle is providing tools that enable the manufacturer to pick the right set of images [to use for fine-tuning] and label them in a consistent way. There’s a very practical problem we’ve seen spanning vision, NLP, and speech, where even human annotators don’t agree on the appropriate label. For big data applications, the common response has been: If the data is noisy, let’s just get a lot of data and the algorithm will average over it. But if you can develop tools that flag where the data’s inconsistent and give you a very targeted way to improve the consistency of the data, that turns out to be a more efficient way to get a high-performing system. “Collecting more data often helps, but if you try to collect more data for everything, that can be a very expensive activity.” —Andrew Ng For example, if you have 10,000 images where 30 images are of one class, and those 30 images are labeled inconsistently, one of the things we do is build tools to draw your attention to the subset of data that’s inconsistent. So you can very quickly relabel those images to be more consistent, and this leads to improvement in performance. Could this focus on high-quality data help with bias in data sets? If you’re able to curate the data more before training? Ng: Very much so. Many researchers have pointed out that biased data is one factor among many leading to biased systems. There have been many thoughtful efforts to engineer the data. At the NeurIPS workshop, Olga Russakovsky gave a really nice talk on this. At the main NeurIPS conference, I also really enjoyed Mary Gray’s presentation, which touched on how data-centric AI is one piece of the solution, but not the entire solution. New tools like Datasheets for Datasets also seem like an important piece of the puzzle. One of the powerful tools that data-centric AI gives us is the ability to engineer a subset of the data. Imagine training a machine-learning system and finding that its performance is okay for most of the data set, but its performance is biased for just a subset of the data. If you try to change the whole neural network architecture to improve the performance on just that subset, it’s quite difficult. But if you can engineer a subset of the data you can address the problem in a much more targeted way. When you talk about engineering the data, what do you mean exactly? Ng: In AI, data cleaning is important, but the way the data has been cleaned has often been in very manual ways. In computer vision, someone may visualize images through a Jupyter notebook and maybe spot the problem, and maybe fix it. But I’m excited about tools that allow you to have a very large data set, tools that draw your attention quickly and efficiently to the subset of data where, say, the labels are noisy. Or to quickly bring your attention to the one class among 100 classes where it would benefit you to collect more data. Collecting more data often helps, but if you try to collect more data for everything, that can be a very expensive activity. For example, I once figured out that a speech-recognition system was performing poorly when there was car noise in the background. Knowing that allowed me to collect more data with car noise in the background, rather than trying to collect more data for everything, which would have been expensive and slow. Back to top What about using synthetic data, is that often a good solution? Ng: I think synthetic data is an important tool in the tool chest of data-centric AI. At the NeurIPS workshop, Anima Anandkumar gave a great talk that touched on synthetic data. I think there are important uses of synthetic data that go beyond just being a preprocessing step for increasing the data set for a learning algorithm. I’d love to see more tools to let developers use synthetic data generation as part of the closed loop of iterative machine learning development. Do you mean that synthetic data would allow you to try the model on more data sets? Ng: Not really. Here’s an example. Let’s say you’re trying to detect defects in a smartphone casing. There are many different types of defects on smartphones. It could be a scratch, a dent, pit marks, discoloration of the material, other types of blemishes. If you train the model and then find through error analysis that it’s doing well overall but it’s performing poorly on pit marks, then synthetic data generation allows you to address the problem in a more targeted way. You could generate more data just for the pit-mark category. “In the consumer software Internet, we could train a handful of machine-learning models to serve a billion users. In manufacturing, you might have 10,000 manufacturers building 10,000 custom AI models.” —Andrew Ng Synthetic data generation is a very powerful tool, but there are many simpler tools that I will often try first. Such as data augmentation, improving labeling consistency, or just asking a factory to collect more data. Back to top To make these issues more concrete, can you walk me through an example? When a company approaches Landing AI and says it has a problem with visual inspection, how do you onboard them and work toward deployment? Ng: When a customer approaches us we usually have a conversation about their inspection problem and look at a few images to verify that the problem is feasible with computer vision. Assuming it is, we ask them to upload the data to the LandingLens platform. We often advise them on the methodology of data-centric AI and help them label the data. One of the foci of Landing AI is to empower manufacturing companies to do the machine learning work themselves. A lot of our work is making sure the software is fast and easy to use. Through the iterative process of machine learning development, we advise customers on things like how to train models on the platform, when and how to improve the labeling of data so the performance of the model improves. Our training and software supports them all the way through deploying the trained model to an edge device in the factory. How do you deal with changing needs? If products change or lighting conditions change in the factory, can the model keep up? Ng: It varies by manufacturer. There is data drift in many contexts. But there are some manufacturers that have been running the same manufacturing line for 20 years now with few changes, so they don’t expect changes in the next five years. Those stable environments make things easier. For other manufacturers, we provide tools to flag when there’s a significant data-drift issue. I find it really important to empower manufacturing customers to correct data, retrain, and update the model. Because if something changes and it’s 3 a.m. in the United States, I want them to be able to adapt their learning algorithm right away to maintain operations. In the consumer software Internet, we could train a handful of machine-learning models to serve a billion users. In manufacturing, you might have 10,000 manufacturers building 10,000 custom AI models. The challenge is, how do you do that without Landing AI having to hire 10,000 machine learning specialists? So you’re saying that to make it scale, you have to empower customers to do a lot of the training and other work. Ng: Yes, exactly! This is an industry-wide problem in AI, not just in manufacturing. Look at health care. Every hospital has its own slightly different format for electronic health records. How can every hospital train its own custom AI model? Expecting every hospital’s IT personnel to invent new neural-network architectures is unrealistic. The only way out of this dilemma is to build tools that empower the customers to build their own models by giving them tools to engineer the data and express their domain knowledge. That’s what Landing AI is executing in computer vision, and the field of AI needs other teams to execute this in other domains. Is there anything else you think it’s important for people to understand about the work you’re doing or the data-centric AI movement? Ng: In the last decade, the biggest shift in AI was a shift to deep learning. I think it’s quite possible that in this decade the biggest shift will be to data-centric AI. With the maturity of today’s neural network architectures, I think for a lot of the practical applications the bottleneck will be whether we can efficiently get the data we need to develop systems that work well. The data-centric AI movement has tremendous energy and momentum across the whole community. I hope more researchers and developers will jump in and work on it. Back to top This article appears in the April 2022 print issue as “Andrew Ng, AI Minimalist.”