Papers accepted at Canada AI and CogSci

We will be presenting 2 papers:

• one at the Canadian Conference on AI (CanAI)
  May 26-29, 2025, in Calgary, Canada, and
• one at the Conference for the Society of Cognitive Science (CogSci)
  July 30 - Aug 2, 2025, San Francisco, USA.

Enhancing Predictive Coding Networks for Multi-Modal Generation and Classification

Ehsan Ganjidoost, Jeff Orchard
To be presented at CanAI.

Abstract: Predictive coding networks (PCnets) provide a biologically inspired framework for classification and generative tasks. However, their generative performance is limited when handling multi-modal class distributions, as the standard weight decay method struggles to differentiate between multiple modes within a class effectively. To address this issue, we propose two extensions: (1) integrating an auto-encoder (AE) to enhance latent representations and (2) employing a predictive coding Hopfield network (PCHN) to capture attractor dynamics. We evaluate these models on synthetic multi-Gaussian datasets and a subset of MNIST, revealing that the AE-enhanced PCnet can generate samples representing distinct clusters. Meanwhile, the PCHN variant independently discovers and maintains cluster structures without explicit latent supervision. Our findings highlight the potential of predictive coding as a sturdy framework for learning multi-modal data distributions in a biologically plausible manner.

A Mechanistic Perspective of Face Perception Latency: Predictive Coding

William Pugsley, Jeremy Zheng, Jeff Orchard, Roxane Itier
To be presented at CogSci.

Abstract: Face processing is widely regarded in cognitive science as the integration of individual features into a holistic percept. However, recent neuroscience research highlights a more nuanced interplay between holistic and featural mechanisms, with specific facial features receiving greater emphasis during early perception. Event-related potential studies reveal that the number and type of parafoveal features significantly influence neural response delays, yet the underlying mechanistic model remains unclear. This paper examines these phenomena through the lens of the predictive coding network, a biologically plausible alternative to traditional deep neural networks. Our findings show that predictive coding networks accurately simulate the influence of parafoveal features on neural response times while upholding the saliency hierarchy of facial features. These results provide a computational explanation for the observed neural delays and highlight the potential of predictive coding as a robust framework for understanding face perception in the human brain.