Training-free spectral fingerprints match related LLM checkpoints from weights
GATE singular-value features stay similar under instruction tuning, quantization, pruning, and distillation without prompts or forward passes.
Stories per day · 30 days
Relatedfeed-forward-networks1inductive-bias1llm-fingerprinting1model-ownership1open-weights1parameterized-quantum-circuits1quantum-machine-learning1radar1
4 stories, newest first · one card per story
Stories tagged #spectral-analysis
Paper links spectral queries to inductive bias in quantum circuits
Encoding a fixed spectral query changes how existing signal frequencies combine without adding sample-dependent information.
Eigenvalues of the Hessian in Deep Learning: The Origin of Symmetry and Its Breaking
A symmetry-breaking account explains clustered Hessian spectra in trained deep networks, including ReLU, CNN, graph, and transformer models.
Predicting Privacy Leakage from Weight Spectral Density
Study shows WeightWatcher spectral metrics like stable rank correlate with membership inference vulnerability, enabling cheaper ML privacy auditing.