Marcus Ghosh

@marcusghosh.bsky.social

973 Followers 382 Following

Computational neuroscientist. https://www.neural-forms.org/

A figure showing a range of neural network architectures. From Ghosh & Goodman, 2025, bioRxiv.
A diagram showing a maze with a gradient (of sensory cues) overlaid.
A diagram showing how different architectures (circles) learn distinct input-sensitivities and memory dynamics.
A diagram comparing the sample efficiency (learning speed) of two architectures (shown in grey and blue), across 4 maze tasks.
A diagram comparing the fitness (task performance) of all pRNN architectures to the fully recurrent architecture, across 4 types of maze environments.
A diagram showing 2D mazes with gradients of sensory cues.
A diagram showing a feedforward network, a fully recurrent network and 3 of the 126, partially recurrent, architectures between these two extremes.
A neural network model with: input, hidden and output nodes, and 9 weight matrices.
A diagram showing 128 neural network architectures.
An image showing five people standing outside of a mint green building (AIMS Cape Town).
A diagram showing an algorithm for evolving neural network models.
A diagram showing model fit (x-axis) vs complexity (y-axis). Evolved models trade-off fit and complexity.
A sample of Python code evolved by the algorithm.
A diagram showing an algorithm which evolves models (in the form of Python code) by combining an LLM with an evolutionary process and some parameter fitting.
A flow chart showing how to compare two reinforcement learning agents.
A four panel figure showing schematic diagrams for linear fusion, nonlinear fusion, nonlinear fusion across short temporal windows and a recurrent neural network model.
A poster advertising the TREND CAMINA summer school.
A diagram of Shannon's setup.
An excerpt from Shannon's presentation (text on page).
The first page of Shannon's transcript (text on page).
An image of Claude Shannon, holding a wooden mouse in a small maze.