NTT starts Physics of Ai Group and AI Design Chip Design for 4K Video

NTT Research has announced at its annual upgrade event that a new AI main research group called has launched a new AI called The physics of the artificial intelligence groupS

Physical AI became a great job in 2025, with Nvidia leading the fee To create synthetic data to pre-control self-driven cars and humanoid robotics so that they can get to the market faster. NTT Research launches its group Physic of Artificial Intelligence (PAI) to get on board.

The new NTT Research independent group is rejected by their Laboratory Physic of Intelligence (Phi) to improve our understanding of AI’s “black box” for better confidence and safety results. NTT Research, which has an annual budget for research and development of $ 3.6 billion, is a subdivision of NTT, the Japanese company for large telecommunications.

Last year, NTT created its vision of Physics of Intelligence, originally formed in collaboration with the Harvard University Center for Brain Science, key contributions made in the last five years, and continued cooperation with academic partners.

The new group will be led by Hidenori Tanaka, a scientific scientist for research on NTT and an expert in physics, neuronuka and machine learning, in a broader pursuit of human cooperation/AI.

The new group will continue to develop an interdisciplinary approach to understand AI introduced by the team in the last five years.

Initially, the Phi Lab has acknowledged the importance of understanding the nature of the “black box” of AI and machine training to develop new systems with drastically improved energy efficiency for calculating. With AI, which is now progressing with amazing speeds, the problems of reliability and safety have also become critical to the applications of the industry and the management of AI.

In collaboration with leading academic researchers, the group of artificial intelligence physics aims to cope with the similarities between biological and artificial intelligence, further understands the complexity of the mechanisms of AI and builds confidence, which leads to a more harmonious merger of human and AI cooperation. The aim is to get a better understanding of how AI works in terms of training, knowledge accumulation and decision -making so that we can design a cohesive, safe and reliable AI in the future.

This approach sounds what physicists have done in many centuries: people have realized that objects are moving when forces applied, but it is physics that has revealed the exact details of the relationship that allows people to design the machines we know today. For example, the development of Steam Engine informs our understanding of thermodynamics, which in turn made it possible to create advanced semiconductors. Similarly, the work of this group will shape the future of AI technology.

The new group will continue to collaborate with the Center for Brain Sciences at the University of Harvard (CBS), led by Harvard Professor Venkatesh Milty, and with an assistant at the University of Princeton (and a former NTT research scientist). He also plans to cooperate with an associate professor at Stanford University Suria Ganguli, with whom Tanaka co -authored several documents. The Group’s main team includes Tanaka, NTT Research Scientist Maya Okawa and NTT Research Doctorate Ekdeep Singh Lubana.

Previous contributions so far include:

• A widely quoted algorithm for arranging a neural network (over 750 quotes in just 4 years)
• An algorithm for the elimination of bias for large linguistic models (LLM) recognized by the National Institute for Standards and Technology of the United States (NIS) for its scientific and practical insights; and
• New insights about the dynamics of how AI learns concepts

In the future, the group “Physics of Artificial Intelligence” has a three -tier mission. 1) He intends to deepen our understanding of the mechanisms of AI, even better to integrate ethics from the inside, not through a fine-tuning patchwork (ie 2) a loan from experimental physics, it will continue to create systematically controllable spaces of AI and monitor the behavior of learning and prognosis of AI step by step. 3) He strives to cure confidence between AI and human operators through improved operations and data control.

“Today it is coming a new step towards understanding AI by society by creating the NTT Research artificial intelligence group,” President and CEO of NTT Kazu Gomi said in a statement. “The Emergence and Rapid Adoption of Ai Solutions Across All Areas of Everyday Life Has Has Had a Profound Impact on Our Relationship with Technology. As Ai’s Role Continues to Grow, IT IT People Feel and How this can the Advancement of New Solutions.

The physics of the group of artificial intelligence covers an interdisciplinary approach to AI, such as physics, neuroscience and psychology gather. This approach seems beyond conventional indicators, recognizing the need to support goals such as justice and safety, which lead to sustainable admission of AI. In terms of energy efficiency, other groups in the PHI lab are already engaged in efforts to reduce the energy consumption of AI computing platforms through optical calculations and road disruption technology, thin -layer lithium niobate (TFLN). On top of that, inspired by the huge difference between watches consumed by LLMS and human or animal brain, the new group will also explore ways to use the similarities between biological brains and artificial neural networks.

“The AI ​​key to exist harmoniously with humanity lies in its reliability and how we approach the design and implementation of AI solutions,” Tanaka says in a statement. “With the advent of this group, we have a way forward to understanding the computing mechanisms of the brain and how it relates to the models of deep training. Looking forward, our study hopes to lead to more natural intelligent algorithms and hardware through our understanding of physics, neuronautics and machine learning.”

Since 2019, the Phi Lab has been running research on new ways for computing systems using photonic -based technologies. TFLN -based devices are explored through this effort, while the ISING coherence machine provides new prospects for complex optimization problems, historically very difficult to solve classic computers.

In addition to the Joint Research Agreement (JRA) with Harvard, Phi Lab has worked over the years at the California Institute of Technology (Caltech), Cornell University, Harvard University, Massachusetts Institute (MIT), Notre University, University of Medica. Overall, the PHI lab has delivered over 150 documents, five appeared in nature, one in science and twenty in nursing magazines of nature.

NTT announces AI output chip for 4K video processing in real time

NTT Corp. Also announced a new large -scale integration (LSI) on AI processing for real -time video conclusions with ultra high definition up to 4K resolution and 30 frames per second (FPS). This low -power technology is intended for final and powerful terminal implementation in which conventional AI inferment requires compression of video with ultra high -time processing definition.

For example, when this LSI is installed on a drone, the drone can detect individuals or objects up to 150 meters (492 feet) above the ground, the statutory maximum altitude of the drone flight in Japan, while conventional AI infection technology in real time will limit that drone operations to about 30 meters (98 feet). One case of use involves the improvement of infrastructure based on drones for operations outside the operator’s visual line, reduction of labor and costs.

“The combination of low -power AI inferment with ultra high -definition video contains a huge
The amount of potential, from infrastructure check to public safety to live sports events, “Gommy said in a statement.” LSI of NTT, which we consider the first of its kind, which achieves such results, is an important step forward in resolving AI’s conclusion in the edge and of the authorities limited by power. “

In terminals limited by edge and power, AI devices are limited to energy consumption, an order of magnitude lower than those of the graphic processors used in AI servers; Dozens of watts of the first compared to hundreds of watts of the second. LSI overcomes these restrictions by implementing an NTT engine AI conclusions. This engine reduces the computing complexity, while guaranteeing the accuracy of detection, improving the efficiency of calculations, using the correlation of the frame and dynamic control of the domestic precision. Performing the object detection algorithm, you only look once (Yolov3) using this LSI, it is possible with energy consumption of less than 20 watts.

NTT plans to commercialize this LSI within the fiscal 2025 through its operating company NTT Innovative Devices Corporation. NTT announced and demonstrated this LSI at Upgrade, the annual meeting of the company’s study and innovation company. Upgrading 2025 takes place in San Francisco April 9-10, 2025.

Looking forward, researchers study the application of this LSI to data -oriented infrastructure (DCI) of the innovative optical and wireless network initiative (IWN), led by NTT and IOWN Global Forum. DCI uses high-speed and low latency capabilities of the IWN All-Photonics Network to cope with the challenges of modern network infrastructure, including obstacles to scale, operating restrictions and high energy consumption.

In addition, NTT researchers collaborate with NTT Data, Inc. Concerning the development of this LSI in connection with its own attribute -based technologies (ABE). ABE enables fine grainy access control and flexible data policy setting, such as shared secretion technologies allow for secure sharing data that can be integrated into existing applications and data stores.

IWN identity

And yesterday NTT announced that Akkira Shimada, President and CEO of NTT, and Kawazoe, Senior Executive Vice President and CTO of NTT, publish a book, IWN identityin which they discuss the IOWN initiative (innovative optical and wireless network), led by NTT, global
Technological leader.

The recently translated book explores NTT’s vision for IOWN and how it will enable a more resilient society in an increasingly managed world.

IWN identity is now available on Amazon after posting during the NTT Annual Summit for Research and Innovation, upgrade. Upgrading 2025 takes place in San Francisco April 9-10, 2025.