Intel is launching research in Israel into machine learning and heterogeneous computing architecture.
The Intel Collaborative Research Institute for Computational Intelligence (ICRI-CI) will focus on long-term research in two directions: machine learning and heterogeneous computing architecture.
Machine learning is critical to transform huge volumes of raw data (e.g., spatial and temporal sensory data, online dynamic data ) into "computational intelligence". Heterogeneous architecture enables performing the required computation fast enough and within acceptable power and area constraints. The combination of new machine learning algorithms and computer architecture that makes these algorithms practical and efficient lay the foundation for many promising and attractive usages together with the needed hardware and software.
The institute will be based in Israel at the Technion in Haifa and the Hebrew University in Jerusalem (HUJI). The Institute is co-led by Ronny Ronen (Intel, Senior PE), by Prof. Uri Weser (Technion), and by Prof. Naftaly Tishby (HUJI). It will also draw researchers from other Israeli universities.
Intel wants to accelerate the research in 3 major themes:
- Advanced Machine Learning. Future devices will use a lot of data arriving from various sources (sensors, web etc) at high rate. Making fast, real time, intelligent decisions requires new type of machine learning algorithms;
- Brain-inspired computing. Humans easily outperform computers in many domains, especially, in learning and recognition tasks. We will apply the deep understanding of brain fundamental structures, principles and mechanisms to explore new, computing architecture that can do these tasks better than traditional computers.
- Novel heterogeneous computing platforms, accelerators. Future usages demand a lot of computing power at tight energy budget to perform tasks like speech and gesture recognition. A promising way of bringing such performance demanding tasks to low power mobile devices is through heterogeneous systems, where building blocks, differing in their capabilities and performance/power characteristics, are combined. The Institute will also investigate the applications of novel machine learning methods in traditional processor architecture to achieve higher performance and efficiency at reasonable complexity.
The institute will apply findings of above fundamental themes to two applications areas and examples of usage scenarios:
- Intelligent Agents. Intel envisions future devices which use machine learning to proactively assist the user's daily activities, based on data coming from "real" sensors as well personal and global data accumulated over a long time via many sources.
- Learning Audio/Visual Systems. Most of the world's data today consists of video and audio streams. The amount of data exceeds the human ability to view and infer from. Intel envisions systems that use machine learning to automatically analyze this data and extract useful relevant information.