arXiv Machine Learning
The Standard Deviation in VICReg Buys a Safety Margin
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
Related stories
arXiv Computer Vision
Aug 25
Bounded Precision-Geometry Scaling for Robust Multi-Task Learning under Loss Scale Mismatch
arXiv:2608.21653v1 Announce Type: cross Abstract: Multi-task learning often combines losses that span several orders of magnitude, causing homoscedastic uncertainty weighting to degrade severely. We...
arXiv Machine Learning
Oct 2
When a Flatness Proxy Is Not a Function: Robustness Certificates and Training Interventions
arXiv:2609.38540v1 Announce Type: new Abstract: A valid curvature upper bound need not justify either a robustness certificate or an intervention on an intrinsic predictor property. We demonstrate th...
arXiv Machine Learning
Jul 7
Directional Curvature from Armijo Backtracking: A Low-Cost Sharpness Probe and a Calibration-Free Learning-Rate Safeguard for Adam
arXiv:2607. 03998v1 Announce Type: new Abstract: The local sharpness of the loss, the top Hessian eigenvalue $\lambda_1$, determines the largest stable gradient step, but measuring it normally requires Lanczos or Hessian-vector iterations.
arXiv AI
Aug 25
Crossing the Margin Cliff: Toward Relearn-Robust LLM Unlearning via Margin Calibration
arXiv:2607.27836v2 Announce Type: replace Abstract: Large language model unlearning is consistently fragile under relearn attacks. On TOFU, fine-tuning on twenty forget examples substantially recover...
arXiv Machine Learning
Jul 13
LionVote: Per-Layer Learning Rate Adaptation for Lion
arXiv:2607. 09266v1 Announce Type: new Abstract: Per-layer diagnostics reveal that, at the prescribed learning rate, Lion's effective scale is 2.
arXiv Machine Learning
Jun 16
False Sense of Safety in Selective Signal Classification: Auditing Bound Tightness and Exchangeability for Risk Control
arXiv:2606. 15153v1 Announce Type: new Abstract: Selective prediction with distribution-free risk control promises that, with confidence 1-delta over the calibration draw, the error rate of accepted inputs stays below a user budget alpha.