Unverified50% confidenceFactExact time
Symmetry breaking refers to a system initially having some symmetry, but during training the symmetry is spontaneously broken, causing neurons to differentiate and take on different functions.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the two-neuron ReLU teacher-student model, small initialization initially makes the two student neurons approximately symmetric; gradient flow first pulls them toward the bisector direction while maintaining symmetry, and the instability of the saddle point subsequently leads to symmetry breaking.66% similarUnverified神经正切核理论由Arthur Jacot等人于2018年提出，其核心洞见是当网络宽度趋于无穷时训练过程退化为核回归问题59% similarUnverifiedThe dynamical attractors before and after a bifurcation have topological structures that are not mutually homeomorphic, which is the fundamental reason the paper names this problem topological out-of-domain generalization.57% similarUnverified双重下降现象的理论解释包括隐式正则化假说、插值阈值理论和神经正切核框架，但尚未形成统一理论框架56% similarUnverifiedJS divergence is a symmetrized version of KL divergence, with results always bounded in [0, ln2], avoiding KL divergence's asymmetry and the defect of diverging to infinity when support sets do not overlap.55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125639API
curl https://kongchang.com/api/v1/knowledge/claims/1125639MCP
get_claim(id=1125639)