Discussion: meetings/2026-02-10

[jeromelecoq]

Here is the updated graphical abstract to discuss tomorrow :

[lrudelt]

Regarding Figure 2, I here post the previous version, and two versions based on what we have discussed in this meeting.

Previous Version

Here, an issue is that not the entire circuit is shown, and the shown circuit is not consistent with panel A.

Proposal 1

In this proposal, I have drawn the complete circuit with positive and negative error neurons in line with panel A. A drawback is that the panel becomes quite busy. Also, this way we do not depict alternative proposals, such as the circuit in Keller & Mrsic-Flogel 2018.

Proposal 2

To solve this, one idea would be to only show one area, because, as John suggest, this contains enough info to map out the entire circuit. If we depict the inter-areal connections sufficiently different from the intra-areal connections, I think the panel becomes better understandable. The idea would be to add another panel as the one shown here under B where a circuit is shown similar to Keller & Mrsic-Flogel 2018 (and the one from the initial draft).

We could do something similar in panel A as well, and make connections in A more abstract, as Martin suggested. However, let's first decide on the scope of the figure and layout, before we invest more work into the drafting.

Let me know what you think.

[smih]

predictionA.pdf

[smih]

predictionB.pdf

[smih]

Hi Lucas,

I think proposal 2 looks beautiful. The comments I had mainly focused on emphasizing what is sent up and down the hierarchy. Simplified cartoons attached above.

[lrudelt]

Hi Stefan,
thank you, I think the cartoons align well with proposal 2 in the sense of depicting inter-areal connections with horizontal lines, and intra-areal with vertical ones. It's also great that you already included the proposal to draw prediction neurons at the bottom and error neurons at the top.

One question to you and everyone else: Do you think it would be sufficient to depict a single area? Because in principle it is clear that if this area repeats how the hierarchical circuit is assembled. This way, we could include an alternative circuit proposal where predictions are sent up (which is then more in line with the proposal in Keller & Mrsic-Flogel). Else, we should consistently include two areas for each row as in your sketches.

[jeromelecoq]

Are you both ready to discuss this at the next meeting ? Anyone else has suggestions/proposais ? *Jérôme*

…

On Fri, Feb 13, 2026 at 12:41 PM Stefan Mihalas ***@***.***> wrote: Hi Lucas, I think proposal 2 looks beautiful. The comments I had mainly focused on emphasizing what is sent up and down the hierarchy. Simplified cartoons attached above. — Reply to this email directly, view it on GitHub <#130 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AATAHTZ46CY4PV6OAAM7XBL4LYZGDAVCNFSM6AAAAACUQOMT6WVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKOBQGAZTENY> . You are receiving this because you authored the thread.Message ID: <AllenNeuralDynamics/openscope-community-predictive-processing/repo-discussions/130/comments/15800327 @github.com>

[lrudelt]

In principle yes, but I think it would be good to discuss beforehand if we want to depict two alternative proposals, or just one with two areas in each row (see above). Then we can already draft an updated version that we can then discuss and finalize on Tuesday.

[smih]

a new version of A
PredictionConA.pdf

[smih]

and a version of c
PredictionC.pdf

[lrudelt]

I tried to incorporate these ideas and draft a proposal for the entire figure