TopoNets: High Performing Vision and Language Models with Brain-Like Topography

2025/01/27 by Mayukh Deb, Mainak Deb, Deb, Mayukh +3 · 16 voices · 8 citations
Computer Science · #Image Retrieval and Classification Techniques #Multimodal Machine Learning Applications #Topic Modeling #cs.LG #cs.NE #q-bio.NC

paper · pdf · doi:10.48550/arxiv.2501.16396

openalex publication_date 2025/01/27 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Neurons in the brain are organized such that nearby cells tend to share similar functions. AI models lack this organization, and past efforts to introduce topography have often led to trade-offs between topography and task performance. In this work, we present TopoLoss, a new loss function that promotes spatially organized topographic representations in AI models without significantly sacrificing task performance. TopoLoss is highly adaptable and can be seamlessly integrated into the training of leading model architectures. We validate our method on both vision (ResNet-18, ResNet-50, ViT) and language models (GPT-Neo-125M, NanoGPT), collectively TopoNets. TopoNets are the highest-performing supervised topographic models to date, exhibiting brain-like properties such as localized feature processing, lower dimensionality, and increased efficiency. TopoNets also predict responses in the brain and replicate the key topographic signatures observed in the brain's visual and language cortices. Together, this work establishes a robust and generalizable framework for integrating topography into leading model architectures, advancing the development of high-performing models that more closely emulate the computational strategies of the human brain.

Cited by

• Model-Guided Microstimulation Steers Primate Visual Behavior
• Beyond binding: from modular to natural vision
• Are Neuro-Inspired Multi-Modal Vision-Language Models Resilient to Membership Inference Privacy Leakage?
• Zero-Shot Performance Prediction for Probabilistic Scaling Laws
• Semantic representations emerge in biologically inspired ensembles of cross-supervising neural networks
• End-to-end Topographic Auditory Models Replicate Signatures of Human Auditory Cortex
• Discovering Functionally Selective Brain Regions with a Deep Topographic Multimodal Model
• Beyond topography: Topographic regularization improves robustness and reshapes representations in convolutional neural networks

Discussions

• TopoNets: High performing vision and language models with brain-like topography [hn, 225 points, 68 comments]
• https://bsky.app/profile/rusirius.bsky.social/post/3lh2kpymsak2f [bsky, 3 points, 0 comments]
• Yes! Here you go! arxiv.org/abs/2501.16396 [bsky, 2 points, 1 comments]
• Intriguing approach to AI model design, mimicking brain structure for enhanced performance. 🤖 #ai TopoNets: High performing vision and language models with brain-like topography [bsky, 2 points, 0 comments]
• TopoNets: High performing vision and language models with brain-like topography (arxiv.org) Main Link | Discussion [bsky, 1 points, 0 comments]
• TopoNets: High performing vision and language models with brain-like topography https://arxiv.org/abs/2501.16396 https://news.ycombinator.com/item?id=42884338 [bsky, 1 points, 0 comments]
• TopoNets: High Performing Vision and Language Models with Brain-Like Topography view on hacker news [bsky, 1 points, 0 comments]
• TopoNets: High performing vision and language models with brain-like topography https://arxiv.org/abs/2501.16396 [comments] [92 points] [bsky, 1 points, 0 comments]
• Inducing brain-like structure in GPT's weights makes them parameter efficient https://arxiv.org/abs/2501.16396 [bsky, 0 points, 0 comments]
• Приведение структуры весов GPT к мозгообразной делает их параметрически эффективными #ai #gpt #news [bsky, 0 points, 0 comments]
• Inducing brain-like structure in GPT's weights makes them parameter efficient https://arxiv.org/abs/2501.16396 (https://news.ycombinator.com/item?id=42884338) [bsky, 0 points, 0 comments]
• Inducing brain-like structure in GPT's weights makes them parameter efficient https://arxiv.org/abs/2501.16396 (https://news.ycombinator.com/item?id=42884338) [bsky, 0 points, 0 comments]
• Inducing brain-like structure in GPT's weights makes them parameter efficient [bsky, 0 points, 0 comments]
• Inducing brain-like structure in GPT's weights makes them parameter efficient #HackerNews arxiv.org/abs/... [bsky, 0 points, 0 comments]
• Inducing brain-like structure in GPT's weights makes them parameter efficient #ai #gpt #news [bsky, 0 points, 0 comments]
• Sleeper paper. This is the kind of work that will give AI big steps forward: arxiv.org/abs/2501.16396 [bsky, 0 points, 0 comments]

Related

• TopoLM: brain-like spatio-functional organization in a topographic language model
• TDSNNs: Competitive Topographic Deep Spiking Neural Networks for Visual Cortex Modeling
• Topoformer: brain-like topographic organization in Transformer language models through spatial querying and reweighting
• Discovering Functionally Selective Brain Regions with a Deep Topographic Multimodal Model
• End-to-end topographic networks as models of cortical map formation and human visual behaviour: moving beyond convolutions