Shulou(Shulou.com)11/24 Report--

The famous British science fiction writer Arthur Clark (2001: space Odyssey) has a saying: "at first glance, any advanced technology is no different from magic." Inside Intel, a huge semiconductor company, there is a group of people who are focusing on this, that is, using novel methods to explore in a wide range of cutting-edge research areas how to help humans meet the major technological challenges in computing, connectivity, cloud-to-edge infrastructure, AI, sensing and sensing.

That's what Intel Research (Intel Labs) is doing. During the 2023 Intel on Technology Innovation Conference, Rich Uhlig, dean of Intel Research Institute, introduced Intel's latest progress and solutions in key technologies such as AI, integrated optoelectronics, neural mimicry computing, quantum computing and so on. He said: "Intel Research Institute's exploration logic is' 3Saving', that is, 'Seek', 'Solve' and 'Scale'. First, try to push the concept forward. Confirm that it can be successful when it enters the large-scale stage, and then establish a technology prototype and test the application of the technology. If successful, it will expand the product scale and turn the research project into a product with the help of Intel's strong execution capability. "

Intel Senior Academician, Vice President and Dean of Intel Research Institute

Rich Uhlig

During this conference, Intel Research Institute also held a variety of activities such as technology insight sharing, thematic forums and Demo demonstrations.

In the technology insight sharing session, Gadi Singer, vice president of Intel Research and research director of emerging AI, shared Intel Research's insights into the future of AI around four aspects: flexibility, cognition, autonomy and responsibility. 2023 is a year of transition to a new stage of AI. The focus of many industries is shifting to large-scale generative AI models that are cost-effective, reliable, safe, responsible and adaptable, including, manufacturing, retail and entertainment. At Intel Research, researchers are promoting emerging AI applications and working with the industry to better integrate and derive value from the latest technologies, helping AI become more flexible, knowledgeable, autonomous and responsible.

In addition, through several special forums, Intel Research Institute introduced in detail its work in many cutting-edge technology areas, including:

Integrated Optoelectronics and its impact on Future Systems and data Center

The increasingly frequent movement of data between servers places a heavy burden on today's data center infrastructure. The industry is rapidly approaching the actual limits of electrical I / O performance. By integrating silicon photons with low-cost, high-capacity silicon, Intel Research Institute's integrated optoelectronic research solves the challenge of growing network traffic in the data center. In this special forum, Intel Research shared its latest progress and introduced the possibility of integrated optoelectronics to change the system interface and network topology in the future data center architecture. it also describes Intel's vision for optical computing interconnection and end-to-end optical networks in future data centers.

AI-driven Edge Service and Control for Intelligent Edge

Enterprises are deploying innovative AI-based or machine learning-based edge services to support various usage scenarios of distributed heterogeneous edges. In this special forum, Intel Research Institute shows how to use AI or machine learning technology to operate the smart edge, from services to end-to-end infrastructure, from top to bottom to platform. The panel introduced a set of innovative edge applications that use AI or machine learning technologies to improve their capabilities and performance, such as those that improve defect detection, perform visual queries and data management, or coordinate the path of mobile robots. Intel Research Institute also shows how automated AI and machine learning algorithm optimization for resource-constrained edge platforms can help optimize application performance, and details how these algorithms can be used to optimize the end-to-end operation of applications on 5G network infrastructure. In addition, at the forum, Intel Research also showed how these algorithms use platform features such as resource scheduling technology (Resource Director Technology) to dynamically adjust the use of shared resources to meet the needs defined by users' overall objectives to implement end-to-end pipelines. it also details how to use innovative AI and machine learning technologies to optimize and operate intelligent edges.

Neural mimicry calculation: Loihi 2 and Lava software

Despite decades of progress and recent breakthroughs, current AI technology still lags behind biological brains in many important areas. In order to create a future world full of intelligent devices and enable them to cooperate with human beings safely and independently, Intel has created a new computing method inspired by the principles of neuroscience. This is the new Loihi processor architecture, which achieves a quantitative increase in computing metrics by minimizing data movement and utilizing transmission based on sparse and asynchronous events. In this special forum, Intel Research Institute introduced its latest progress in the field of neural mimicry computing in detail.

Promote large-scale basic AI research

Intel's AI Lab develops basic AI technologies, disruptive applications, and challenging programs. The insights shared at the forum involved the gigabyte Neural Network (GNN) distributed training platform, the understanding of long videos through graphical architecture (processing 10x long videos while matching the performance of the SOTA conversion model at very low cost) and "AI for Science" breakthroughs (the first dialogue interface for protein structure design and a new benchmark for AI for new material research and development).

The Progress of Quantum Computing and the introduction of Intel ®Quantum Software Development Kit (Intel ®Quantum SDK)

Although it may take many more years, quantum computing is expected to make breakthroughs in energy production, materials, chemical and drug design, financial and climate modeling, and cryptography. Many companies are working together to promote research that contributes to the realization of quantum computing. Intel Research is expanding the software developer community to take full advantage of a full-stack solution called Intel ®Quantum Software Development Kit. In this forum, Intel Research Institute detailed the latest progress made by Intel, introduced and demonstrated how to use Intel ®Quantum Software Development Kit, including running popular classical quantum chemistry algorithms.

Computing: exploring the Frontier of data Privacy Protection

Beyond today's secret computing, a new research field called computing is expected to further improve data privacy and security in cloud computing. Unlike the trusted execution environment, computing is a powerful new technology that implements end-to-end computing and collaboration involving private sensitive data. It may revolutionize the way data is shared, analyzed, and processed in the future, enabling organizations to gain valuable insights while theoretically eliminating the risk of data exposure to third parties. In this special forum, Intel Research Institute introduced the latest progress made in homomorphism and Intel computing software development kit.

Self-optimizing Database: using Machine Learning in Database Management system

Database management system (DBMS) is a complex software system. Its performance depends on hardware, data, and database queries. Optimizing DBMS for specific problems can be overwhelming. Intel Research Institute, working closely with academic partners of MIT and Brown University, is developing a data management system that automatically completes self-optimization. DBMS optimization using simple parameter search methods has been used before, however, this research is opening up new directions through innovative methods based on machine learning. From self-organizing data containers and auto-optimized query engines to low-level data structures based on machine learning models in DBMS, everything is clear. In this special forum, Intel Research Institute introduced the data management of self-optimizing system in the future.

Heterogeneous programming: distributed data structures, algorithms and views in C++

Distributed and heterogeneous systems, such as clusters with multiple GPU per node, have become more and more common. At present, the programming of such systems usually uses multiple low-level programming models. In order to build a software ecosystem for these systems, a more advanced programming model is needed. The goal of Intel Research is to help users use standard data structures and algorithms in the programming of such systems, just as they currently use sequential systems. Intel Research uses a distributed data structure that automatically distributes data across multiple GPU or nodes. In addition, the distributed algorithm of parallel computing is also used. In this special forum, Intel Research introduced how to run C++ programs on multiple GPU with minimal modification to make performance comparable to professionally tuned code.

At the same time, Intel Research Institute also displayed a number of Demo at the 2023 Intel on Technology Innovation Conference, showing many explorations and attempts to promote the application of cutting-edge technology, including:

AI Robot: a Programmable Human-AI interaction Technology

Cooperative robot (Cobot) is becoming an adaptive intelligent assistant for general manual labor, with a subversive return on investment. In this demonstration, the end user controls the robot (remotely) through VR to directly complete the task of picking up and placing items. The programming process shown includes imitating humans, marking movements and entities, and arranging the skills of modular robots. This usage scenario gradually demonstrates and introduces Intel's progress in motion primitives, enhanced robot vision and VR interfaces, controlling, reasoning, determining, and acting in real time through Intel RealSense technology and Intel XPU hardware.

AI security and trust

This demo covers several topics such as security, trust, and privacy protection in the AI domain. While popular large language models (LLM), such as ChatGPT, are trained with publicly available data, the next step is to deploy them within the organization and further enhance them with custom data. But how do you ensure that the large language model respects organizational boundaries and does not disclose information to unauthorized people? Interactive Demo presents a guessing challenge aimed at persuading the big language model to reveal what it is told not to disclose.

Auxiliary calculation: AI reply generation and brain-computer interface

The Auxiliary context Awareness Toolkit (Assistive Context-Aware Toolkit, ACAT) is an open source software platform developed by Intel Research that enables people with motor neuron disease to communicate through keyboard simulation, text prediction and speech synthesis. The recently released version 2. 0 of ACAT has a brain-computer interface that allows users who can't move at all to communicate. This version also includes enhanced language modeling capabilities that are fine-tuned for AAC (enhanced and alternative communication) uses to achieve personalized sentence completion, as well as more efficient typing modes and faster communication. The system is optimized to run on client devices, while the completely redesigned user interface is easy to use. This demo shows all the features of ACAT, including the brain-computer interface and the new text prediction function.

Generative AI:3D potential diffusion model (LDM3D)

DepthFusion demonstrates the power of Intel's LDM3D diffusion model, generating 360-degree views through text prompts. The LDM3D diffusion model uses user-provided text prompts to generate 2D RGB images and their corresponding depth maps, thus providing a complete RGBD rendering of text prompts. Then, the model is fine-tuned using a subset of the LAION400M data set, which is a large-scale graph-text data set. The depth map used to fine-tune the model is generated by MiDaS 3.1, a depth estimation algorithm that provides a highly accurate relative depth estimation for each pixel in the image. Intel applications then take advantage of the power of the multi-functional platform TouchDesigner to create an immersive interactive multimedia experience, generate realistic 360-degree views, and provide users with a unique and fascinating way to experience the text prompts they give.

Immersive remote presentation on Light Field display

The Demo focuses on a complete real-time Lightfield video end-to-end system that uses cloud processing and is supported by client functions in terms of content creation and consumption of end devices. This "Lightfield Video" technology demonstrates breakthroughs in immersive video-on-demand, live streaming and videoconferencing experiences. The demo shows a complete end-to-end system, including a consumer light-field camera array for simultaneously capturing a variety of camera angles, algorithms for variable viewpoint video creation, cloud processing, and streaming content delivery. and an immersive video viewing device.

Neural object replication: AI-driven 3D content creation for games and other fields

Neural mimicry Computation: satellite scheduling and Optimization using Loihi 2

The demonstration shows how Intel's neural mimicry technology can reduce the computing burden when solving scheduling problems often encountered in logistics and operations. A typical scheduling problem involves assigning a certain number of agents to perform a specific number of tasks while complying with scheduling constraints. The demonstration shows how to solve a key scheduling problem in the space technology industry: assigning a large number of Earth observation requests (tasks) to satellite clusters (agents). The time taken by the scheduling problem is proportional to the number of agents and the square of the number of tasks to be solved. For commercial satellite companies with dozens of satellites and thousands of customer requests, the current algorithm can not find the best solution in time.

Quantum computing: Intel ®Quantum Software Development Kit and 3D interactive hardware demonstration

The demo highlights version 1. 0 of the Intel ®Quantum SDK, which allows users to dock Intel's quantum computing stack. The software development kit includes an intuitive C++-based user interface, a LLVM-based compiler tool chain for quantum computing, and a high-performance Intel Quantum Simulator (Intel Quantum Simulator) as the back end of qubit target code. The demo also shows the 3D interactive look of Intel Quantum hardware.

Towards the future of computing, mankind will not stop. Intel Research Institute will always be committed to "all-round attack" in various cutting-edge areas, from technological exploration to application landing, constantly exploring diversified possibilities and "planting wings" for future computing.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.