Likelihood Approximations for Cognitive Modeling with PyMC

Dive into cognitive modeling with PyMC. Learn about the impact of likelihood approximations.


AUTHORED BY

Thomas Wiecki

DATE

2023-01-12


Introduction

Digital therapeutics are evidence-based, clinically evaluated software and devices that can be used to treat an array of diseases and disorders, according to the Digital Therapeutics Alliance, the industry's trade association. They can be used independently or with medications, devices, and other therapies to treat physical and behavioral health conditions, including pain, diabetes, anxiety, post-traumatic stress disorder, and asthma.

In this talk, PyMC Labs and Akili discuss using Bayesian methods and PyMC to test a range of computational models of cognition, specifically with an eye towards ADHD (Attention-deficit/hyperactivity disorder). They focus on some technical challenges and how ideas from likelihood-free inference and machine learning can help overcome them.

Timestamps

00:00 Thomas Wiecki introduction

01:27 Alex introduction

01:51 Titi introduction

02:55 Andy introduction

03:42 Akili background

05:11 EndeavorRx by Akili and PyMC's involvement

09:49 Likelihood approximation networks in PyMC

10:15 NeuroRacer

15:44 Two important aspects of the Model

20:39 Inference with model variants

21:05 Inference from access to simulators

21:55 Inference with models

22:56 Training

24:06 Previous toolbox: HDDM

24:59 Properties inherited from Neural Networks

26:31 Graphical representation of model in PyMC

29:42 Code in PyMC

30:07 Neural network (LAN, CPN)

31:17 Proof of concept (Parameter Recovery)

31:42 Proof of concept (Speed)

32:41 Thomas question on speed

36:11 Thomas on Before PyMC vs after PyMC

36:32 Titi on before PyMC vs after PyMC

38:11 Andy on production use case

39:00 Thomas question on application

39:16 Andy explains the use case and the application

40:28 Titi on impact in applications

41:15 Thomas on Knowledge transfer to Akili research team and collaboration

43:02 Andy on working with PyMC team

44:55 Thomas question to Alex on applying this method to other applications across industries

47:15 Why does Akili care about these kinds of models ?

49:31 PyMC's work and impact towards Akili's mission

51:04 Audience Q/A (What other conditions can this be applied other than ADHD?)

52:03 Audience Q/A (Is Data enough to conduct experiments ?)

56:32 Closed form solution vs Neural Networks

56:52 Optimizing LAN for faster forward pass, primary metric and designing networks

Resources


Work with PyMC Labs

If you are interested in seeing what we at PyMC Labs can do for you, then please email info@pymc-labs.com. We work with companies at a variety of scales and with varying levels of existing modeling capacity. We also run corporate workshop training events and can provide sessions ranging from introduction to Bayes to more advanced topics.