State Space Models: Advancements and Applications in AI - Featuring Albert Gu and Karan Goel

Overview

Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!

Grab it

Dive into a comprehensive video lecture on state space models featuring Albert Gu and Karan Goel from Cartesia AI. Explore the cutting-edge research in SSMs competing against transformer supremacy, including insights from the seminal S4 paper and the Mamba paper. Learn about real-time intelligence, text-to-speech demos, and the fundamental concepts of state space models. Examine the trade-offs between SSMs and transformers, understand how SSMs are computed convolutionally, and compare different state representations in machine learning models. Discover the latest developments in Mamba 2 and discuss scaling laws in AI. Gain valuable knowledge from these pioneering researchers in the field of state space models and their applications in artificial intelligence.

Syllabus

00:00:00 - Intro
00:01:56 - Hyperstack GPUs! sponsored
00:03:37 - Cartesia AI, real-time intelligence
00:09:45 - text-to-speech demo
00:12:05 - state space models SSMs
00:36:35 - bitter lessons transformers vs SSMs
00:38:30 - free lunch? what are the actual tradeoffs?
00:41:30 - Computing SSMs convolutionally
00:42:55 - What is the state RNNs, Attention, Conv, Neural ODEs
00:50:00 - Mamba 2, discrete vs continuous spectrum
00:55:30 - Q&A: scaling laws