Ryan Smith ~ Active Inference Insights 013 ~ Wellbeing, Modelling, Precision
May 3, 2024
auto_awesome
Neuro-computationalist Ryan Smith discusses empirical computational psychiatry, in-silico simulations, and the role of aberrant precision-weighting in neuropsychological disorders. The podcast delves into optimizing well-being through free energy, decision uncertainty in mental health conditions, exposure therapy, preferences in active inference models, emotional granularity, and future work in the field.
Understanding the difference between empirical computational psychiatry and in-silico simulations is crucial for modeling neuropsychological disorders.
The interplay between variational and expected free energy influences decision-making processes and action dynamics.
Model accuracy in estimating expected free energy is essential for making precise policy decisions and adjusting learning rates.
Deep dives
The Complexity of Subjective Well-Being and Active Inference Mechanisms
The discussion delves into the complexity of grounding subjective well-being in the parameters and mechanisms of active inference. While variational free energy aims to optimize posterior beliefs about states, expected free energy focuses on rewarding outcomes and information gain. The distinction highlights valency considerations for subjective well-being beyond basic biological fitness.
Action Dynamics in Variational and Expected Free Energy Minimization
The episode explores the interplay between variational and expected free energy in action dynamics. Variational free energy serves to update beliefs about states based on observations, while expected free energy factors in rewardingness of outcomes and influences decision-making processes. The discussion also touches upon combining expected and variational free energy in decision-making contexts.
The Role of Model Accuracy in Learning Rates and Expected Free Energy
The conversation delves into the reliance on model accuracy in adjusting learning rates and estimating expected free energy. The accuracy of calculating expected free energy is contingent on precise posterior beliefs over states and appropriate model updates, impacting decision-making processes. The episode also touches on the influence of habits in decision-making under uncertainty.
Empirical Computational Psychiatry and Model Comparison for Behavioral Data
The discussion transitions to empirical computational psychiatry and fitting models to behavioral data in clinical populations. By comparing the effectiveness of active inference models to more traditional reinforcement learning models, insights are gained into decision-making behavior. The study emphasizes the importance of model comparison to assess accuracy and predictive capabilities of active inference models.
Importance of Precise Information in Generative Models
Fine-grained state space carving enhances the informativeness of generative models leading to informed policy selection. Inferring accurate emotional states and unique predictions under each state bolsters policy decision-making. Structure learning and parameter learning distinction require specific implementations for precise inference and policy selections.
Empirical Examination of Predictive Coding and Active Inference
Active inference and predictive coding frameworks necessitate empirical validation through model comparisons to existing theories. Testing on various datasets reveal active inference models outperforming traditional models in explaining two-step task behaviors. Further research aims to refine hypotheses and testing methods to provide conclusive evidence supporting these innovative frameworks.
Active Inference Insights 013 is all about modelling: not only how we model our lived world, but also how we might scientifically model that modelling process ourselves. With neuro-computationalist Ryan Smith at the helm, expect to learn about the difference between empirical computational psychiatry and in-silico simulations, the role of aberrant precision-weighting in neuropsychological disorders and how a little parameter called ‘gamma’ might hold the key to a life of wellbeing.
Get the Snipd podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode
Save any moment
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Share & Export
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode