Panel Presentation: Liquid Neural Networks with Live Q&A

Presented at: Neuroscience Virtual Event Series 2023

3 +10pts

Speakers

Mathias Lechner, PhD

Postdoctoral Research Associate, Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology (MIT)
BIOGRAPHY
Ramin Hasani, PhD

Principal AI and Machine Learning Scientist, Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology (MIT)
BIOGRAPHY

Abstract

In this talk, we will discuss the nuts and bolts of the novel continuous-time neural network models: Liquid Time-Constant Networks (LTCs). Liquid neural networks are nonlinear state-space models with an input-dependent varying (i.e., liquid) time characteristic. Their outputs are computed by numerical differential equation (DE) solvers when described by ordinary DEs, and by continuous functions when described in closed-form. Liquid networks exhibit stable, bounded and dynamic causal behavior, yield superior expressivity within the family of CT neural models, and give rise to improved performance on a wide range of time series prediction tasks In and out of distribution compared to advanced recurrent models.

Learning Objectives:

1. Discuss the merits and pitfalls of The Awake Closed Head Injury (ACHI) model for brain injury research.

2. Describe how mild traumatic brain injury can affect the structure and function of the brain.

3. Discuss opportunities for future studies using a clinically relevant model of mild traumatic brain injury like the ACHI.