Meeting Notes

• Date: 2026-02-17
• Time: 09:00AM (PT)
• Location: Teams Meeting

Agenda

Discussion on figure update

• @smih and @lrudelt will present sketch proposals for figure 2 Check the previous meeting notes to follow those discussions

Meeting Recording

Meeting Notes

Discussion of Predictive Coding Models and Literature References: Stefan, Jerome, and Lucas Rudelt discussed various predictive coding models, including differences between hierarchical and non-hierarchical error propagation, and referenced recent literature such as Keller's and Whittington's work, with Stefan clarifying distinctions and similarities among these models.

Model Comparison: Stefan explained the differences between the model he referenced and Keller's model, noting that the former features bottom-up and top-down error neurons, while Keller's model involves error propagation in both directions but with different organizational pathways.

Recent Literature: Jerome and Lucas Rudelt mentioned recent bioRxiv and machine learning literature, including JEPA and cognitive science studies, highlighting their relevance to the group's ongoing work and suggesting these as useful references for figure development.

Clarification of Whittington's Model: Stefan clarified that Whittington's model, as discussed by Rafa, operates predictive coding in two modes—an inferential mode and a clamping mode for plasticity—without fundamentally modifying predictive coding, but rather switching operational modes.

Citations and Attribution: The group discussed the challenge of citing all relevant literature, with Jerome and Stefan acknowledging that not all models or concepts are always directly cited in their work, but that the ideas are often discussed in various sections.

Presentation and Iterative Revision of Figure 2: Stefan and Lucas Rudelt presented updated proposals for Figure 2, incorporating feedback from previous meetings, and the group, including Farzaneh, Jerome, and others, provided detailed suggestions for improving clarity, labeling, and conceptual representation of hierarchical and non-hierarchical circuits.

Panel Structure and Conceptual Focus: Stefan described the revised structure for Figure 2, emphasizing clearer visual separation between within-area and between-area connections, and proposed making the panels more conceptual to highlight the flow of information rather than depicting detailed synaptic structures.

Microcircuit Representation: Lucas Rudelt explained the depiction of microcircuits in the updated figure, distinguishing between hierarchical and non-hierarchical organizations, and discussed the challenges in naming conventions and the representation of error and prediction signals.

Simplification and Labeling Suggestions: Farzaneh suggested breaking down complex panels (C and D) into simpler components, possibly by separating positive and negative error circuits, and advocated for concise summary phrases or numbered steps next to each panel to guide readers through the figure.

Color Coding and Accessibility: The group discussed color schemes for representing positive, negative, and unsigned errors, with Stefan proposing yellow, red, and orange, and Jerome emphasizing the need to consider color blindness and to consult Benedicte for final color choices.

Naming Conventions and Consistency: Stefan and Lucas Rudelt agreed to use terminology consistent with Keller's and other referenced papers for hierarchical and non-hierarchical circuits, ensuring that the figure's labels match established literature.

Technical Discussion on Error Computation and Circuit Details: Martin, Lucas Rudelt, Stefan, and Jerome engaged in a technical discussion about the physiological and computational interpretation of error computation in dendrites, the role of inhibitory neurons, and the distinction between representation and prediction in the figure panels.

Dendritic Error Computation: The group debated how to represent error computation in dendrites, with Martin and Lucas Rudelt discussing whether to depict positive and negative error dendrites explicitly and how these relate to physiological reality.

Inhibitory Neuron Representation: There was discussion about whether to include feedforward or recurrent inhibitory neurons in the figure, with Lucas Rudelt suggesting to only depict recurrent inhibition on basal dendrites and to remove ambiguous connections for clarity.

Representation Versus Prediction: Martin and Lucas Rudelt clarified that feedforward signals should be labeled as representations rather than predictions, as predictions are generated locally via lateral inhibition, and this distinction should be reflected in the figure and its caption.

Color and Labeling for Error Signals: The group agreed to use specific colors for positive, negative, and unsigned errors, and to ensure that the legend and panel labels accurately reflect the computational roles of each circuit component.

Action Items and Next Steps for Figure Revision: Lucas Rudelt agreed to incorporate the group's feedback into a revised version of Figure 2, post the update on GitHub for further review, and the team, led by Jerome, planned to revisit the figure and discuss analysis tables in the next meeting.

Figure Revision Plan: Lucas Rudelt committed to updating Figure 2 based on the discussion, including changes to color coding, labeling, and circuit representation, and to share the revised figure on GitHub for team feedback.

Future Meeting Agenda: Jerome outlined the agenda for the next meeting, which will include reviewing the updated figure and introducing analysis tables to help the team track data modalities and analysis readiness.

Collaboration and Feedback Process: The team agreed to continue the iterative process of figure revision, with Benedicte assisting on layout and accessibility, and to use GitHub discussions for asynchronous feedback before the next meeting.