NeurIPS 2026 reviews are up: reviews, scores, rebuttals and meta-reviews for 7,877 papers. See the highest rated
acceptodds: the papers of ICLR 2027
138 results for “keyword:"training dynamics"” in ICLR 2027. 35 withdrawn or desk rejected listed apart. clearsort:relevanceratingnewestmost read
[PDF]Attention Sinks in Transformer-based Language Models: Emergence Mechanisms and Benign Effects▲5.4k3.00
[PDF]Disentangling Regularization Effects in Self-Supervised Learning through Controlled Gradient Interventions▲6.1k2.00
[PDF]Is your LLM a sequence model on the training history? The origins and consequences of anticipation▲9.2k2.67
[PDF]Stagewise Learning of Balanced Orthogonal Mixtures by Finite-Width Two-Head Transformers▼1361.33
[PDF]Initial Spectral Power Predicts Frequency Selection in Grokking and Steers It in Tied MLPs▼1361.33
[PDF]Enhanced Generation in Diffusion Models via Joint High-to-Low Denoising Schedule and Model Sparsity with Theoretical Guarantees▲3.0k2.33
[PDF]The Life Cycle of a Massive Activation: Stochastic Birth, Weight-Decay-Driven Growth, and Competitive Consolidation▼4.3k2.00
[PDF]Theoretical Analysis of Multimodal Contrastive Learning: When and How Hard Negatives Help▼4.3k2.00
[PDF]Fighting Fire with Fire: Assessing Test Set Contamination via Deliberate Training on Test Data▲3.0k2.33
[PDF]Collapse, Repair, and Drift: Dissecting Fine Tuning Dynamics in Speech Language Models▼1.2k1.67
[PDF]Understanding Optimization from Activation and Gradient Spectra: A Modded-NanoGPT Study▼13k2.00
[PDF]Normalization as Control: Batch Normalization Is the Deadbeat Corner of a Kalman-Tuned Normalizer Family▼4.3k2.00
[PDF]Same Fit, Different Functions: Evidence Structure and Training Order Shape Held-Out Behavior▲11.00
Withdrawn or desk rejected (35)
[PDF]Tracing How Backdoors Are Learned: Poisoned Sample Detection via Information-Trajectory DistortionWithdrawn1.33
[PDF]Data-Weight Directions Rotate Along Training: Horizon-Matched Attribution for GUI TrajectoriesWithdrawn1.33
[PDF]Learning to Be Fooled: pretraining grows knowledge but keeps the scar of framing susceptibilityDesk rejected
[PDF]Before the Breakthrough: Bottleneck Crossings as Precursors in Long-Horizon Agent RLWithdrawn2.00