Unverified50% confidenceFactExact time
When a layer of Inception V1 was cut from 480 neurons to 10 and retrained, the network did not exhibit superposition; instead, neuron functions became coarser and more generalized.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Neural Networks Were Always Clean? Goodfire Researcher Overturns Superposition Theory
bilibili迦太基盐商10/16/2026
Related Entities
Related Claims
UnverifiedCammarata's extreme stress experiment showed that after compressing 480 curve-detecting neurons in a certain layer of InceptionV1 down to 10 and retraining, these neurons were still clearly interpretable rather than chaotically superposed.76% similarUnverified叠加现象指神经网络倾向于用远少于特征数量的神经元同时编码大量语义特征，与传统一个神经元一个特征的直觉相悖73% similarUnverifiedFor each generator, the key neurons consistently selected account for less than 1% of the total, and they are highly consistent across different cross-validation folds and random seeds.72% similarUnverified深度神经网络天然倾向于过拟合训练分布，对分布外样本的处理能力往往大幅下降69% similarUnverifiedThe precise calculation of Φ faces a combinatorial explosion problem that already exceeds the capacity of current computers even for networks of tens of nodes, while the human brain has 86 billion neurons.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1052470API
curl https://kongchang.com/api/v1/knowledge/claims/1052470MCP
get_claim(id=1052470)