February 16th, 12pm EST
Dr. Nir Shavit is a Professor at MIT, a member of the Computer Science and Artificial Intelligence Laboratory, known for his work in multi-core computation.
Talk Highlights: Tissue vs. Silicone Musings on the Future of Deep Learning Hardware and Software
Dr. Shavit explores the relationship between neurobiology and the design of hardware and software for machine learning. He demonstrates that the human brain operates at roughly iPhone-level compute but on a petabyte of memory — the opposite of current GPUs. He argues that the future lies in mimicking the brain's sparsity and locality of reference rather than simply adding more floating-point operations.
Key Takeaways
- The human brain performs roughly 2 trillion operations per second — comparable to an iPhone — not a massive TPU pod. We are building the opposite of what we need.
- GPUs are built with massive compute and tiny memory, but to mimic a brain we need iPhone-level compute paired with a petabyte of memory.
- By applying pruning and cache-aware depth-wise execution on CPUs, researchers can match GPU latency for inference tasks like BERT — brain-inspired software strategies can close the hardware gap.
Notable Quotes
“If you want to mimic us, you don't need that much compute. An iPhone does five trillion operations per second. My brain is not more powerful than an iPhone.”
“We don't need all those flops. What we need is memory and the right algorithms.”
Biography +
Nir Shavit
Early Life and Education
Nir Shavit, born in 1959 in Israel, is a computer scientist and leading authority on concurrent and parallel computing. He earned his B.S. (1984) and M.S. (1986) from the Technion, and his Ph.D. from the Hebrew University of Jerusalem in 1990.
Career and Contributions
Shavit held a professorship at Tel Aviv University before joining MIT in 2011. With Maurice Herlihy and others, he applied algebraic topology to shared-memory computability, earning the Godel Prize (2004). He co-introduced software transactional memory (STM), recognized with the Dijkstra Prize (2012). He co-authored "The Art of Multiprocessor Programming" (2008) with Herlihy. He co-founded Neural Magic, acquired by Red Hat in 2024.
Awards and Honors
- 2004: Godel Prize
- 2012: Dijkstra Prize
- 2013: Named ACM Fellow
- Present: Professor at MIT EECS
Career Timeline +
Career Timeline
- 1959: Born in Israel
- 1984: B.Sc. from the Technion
- 1986: M.Sc. from the Technion
- 1990: Ph.D. from Hebrew University of Jerusalem
- 1990s: Professor at Tel Aviv University
- 2004: Godel Prize for shared-memory computability work
- 2008: Published "The Art of Multiprocessor Programming"
- 2011: Joined MIT
- 2012: Dijkstra Prize for software transactional memory
- 2013: Named ACM Fellow
- 2024: Neural Magic acquired by Red Hat
