Microsoft search forum (Opens in a new tab) is a new series of conversations that explore recent advances, daring new ideas and important discussions within the global research community. Microsoft’s main researchers will share information on their work, followed by online discussions live with public participants.
This article gives an overview of the inaugural conversation of the Microsoft search forum, with a summary of each presentation. All details, including Co -pilot’s experience (Opens in a new tab) and the reruns of each session (Opens in a new tab)are available on demand. Register (Opens in a new tab) To attend to come to the coming events of the research forum.
Peter Lee, CVP, Microsoft Research & Incubations
2023 was an incredible year for AI research, with rapid change and the emerging sparks of artificial general intelligence. Generative AI now influences everything in research, and research has never more important the innovative technology that will benefit society. And although there are many reasons of optimism, we must also be clear about risks and limitations – another direction where research can play an important role.
In this environment, openness and collaboration are essential, not only to advance research, but to ensure that technology is developed with a commitment to security and ethical use. Microsoft continues to invest in its commitment to responsible AI (RAI), which is deeply integrated not only into each engineering group through the company, but also through functions such as finance, security and Legal teams. Additional progress will require close collaboration with the wider research community.
Some of the most promising and tangible advances present in medicine and materials science. The examples include Microsoft AI4Science’s work, a Microsoft research laboratory, which works with the Global Health Drug Discovery Institute to accelerate the discovery of new treatments for infectious diseases.
Ashley Llorens, vice-president and distinguished scientist, Microsoft
ECE Kamar, Managing Director, Microsoft Research AI Frontiers
Sébastien Bubeck, VP, Microsoft Genai
Ahmed Awadallah, Senior Main Research Director, Microsoft Research AI Frontiers
The panelists explored their aspirations at AI in the near future, as well as the challenges to be overcome. Examples include:
- Go beyond language to build AI systems that become aid in the physical world. AI can more than answer questions; He can better understand our goals and intentions and create a difference in people’s lives.
- Beyond trying to obtain AI to imitate the human mind, can we really light up How does the human mind work and discover the constituent elements of reasoning?
- Make IA technology smaller would help reduce cost and increase the performance of current AI systems. How to divide the problems into small pieces to solve? And how can we reduce the requirements of megadata, large networks of neural and massive IT resources?
- Can we create a virtuous feedback loop, where AI learns people who use them, rather than simply providing answers from a static basis of information?
The panelists also explored the rapid pace of technological development. Historical deadlines of three to five years are now condensed in a few weeks. In this environment, collaboration is essential to quickly develop ideas and increase experimentation between organizations. This also amplifies existing concerns concerning the optimization of safety and attenuation of biases in language models.
Lightning talks
Improvement of reasoning in language models with laser: reduction of selective layers of layers
DiPendra Misra, principal researcher, Microsoft Research NYC and AI Frontiers
Large language models (LLM) have revolutionized automatic learning. While researchers continue to advance this technology, an approach is to do an intervention in models and observe how it affects their performance. This conference presents laser, a new intervention method that can increase the accuracy of LLMS while reducing their memory footprint.
Evaluation and understanding of foundation models
Besmira Nushi, principal researcher, Microsoft Research AI Frontiers
The evaluation and understanding of the models serve as guides for IA innovation. But the evaluation is difficult and the new generative tasks take new challenges in evaluation and understanding. This presentation explores efforts to measure, inform and accelerate the improvement of the model, which helps the scientific community to understand and study new forms and levels of intelligence.
Generative AI meets structural biology: prediction of equilibrium distribution
Shuxin Zheng, principal researcher, Microsoft Research AI4Science
The Distributional Graphormer (DIG) is an in -depth learning framework to predict protein structures with greater precision, a fundamental challenge in molecular science. Using a generative AI to resolve the problem of predicting balance distribution, DIG opens up new fascinating possibilities. By learning different states and behaviors of molecules, scientists can make breakthroughs in the development of new drugs, the creation of advanced materials and understanding biological processes.
Increase human cognition and decision -making with AI
Jake Hofman, principal principal researcher, Microsoft Research NYC
How can AI help people make better decisions, be more productive and improve in a sustainable way? Some technologies can help in the short term without providing sustainable solutions. For example, relying on a spelling verifier may not improve their ability to spell properly. This conference explores the choices in the design and use of AI tools to help decision -making and the importance of a measure and an rigorous experiment to maximize the advantages and minimize the risks.
Kahani: visual narration through culturally nuanced images
Sameer Segal, main research software development engineer, Microsoft Research India
Image generation models can produce visually breathtaking images from natural language descriptions, but they often lack cultural consciousness and nuances. These models can rely on stereotypes and not understand the local words, which require heavy fixes such as modification or considerably end of the model. The generation of images can also require sophisticated incentive, beyond the capacities of many lay people.
This conference examines Kahani, a Microsoft research project focused on the development of a visual narrative prototype that allows people to create visually striking images and culturally nuanced simply by describing them in their local language. Kahani uses advanced techniques such as detention and models such as all and GPT-4V segment (ISION) to generate comments for candidate images.
Ashley Llorens, vice-president and distinguished scientist, Microsoft
The acceleration of AI highlights the importance of commitment between disciplines, organizations and geographies. This session presented the first cohort of scholarship holders to Microsoft Research’s Ai & Society Fellows (Opens in a new tab) Program, which aims to promote a deep interdisciplinary collaboration which maximizes the value of the AI for people and society. The session also provided an update on the Accelerate research on foundation models (Opens in a new tab) (AFMR) Program, which offers subsidies that make leading models, animated via Microsoft Azure, accessible to university research teams. To date, AFMR subsidies support nearly 200 projects in 80 research institutions around the world. These projects include work in the innovation and evaluation of the AI model, responsible AI, health, AI for scientific discovery, etc.