{"total":11,"items":[{"citing_arxiv_id":"2605.19343","ref_index":67,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"What Makes a Representation Good for Single-Cell Perturbation Prediction?","primary_cat":"cs.LG","submitted_at":"2026-05-19T04:30:11+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"PerturbedVAE disentangles perturbation-specific signals from invariant gene expression structure to recover causal representations and improve out-of-distribution prediction in single-cell perturbation modeling.","context_count":1,"top_context_role":"background","top_context_polarity":"unclear","context_text":"Since identifiability is only defined up to an unknown transfor- mation at the block level, we fit a regression model from the learned latents to the ground-truth latents. Concretely, we fit a functionfof the form zι =f( ˆzι) +ϵ.(66) 3.Compute the coefficient of determination.After fitting the regressor, we compute the coefficient of determination R2 = 1− Pn i=1 ∥zι,i − ˆz pred ι,i ∥2 Pn i=1 ∥zι,i − ¯zι∥2 ,(67) where ˆz pred ι,i =f( ˆzι,i)is the predicted ground-truth latent for samplei, ¯zι is the empirical mean of the ground-truth latents, andnis the number of test samples. 4.Interpretation.A value ofR 2 close to1indicates that the learned latent block can be (nonlinearly) transformed to accurately recover the true latent block, consistent with block-wise identifiability."},{"citing_arxiv_id":"2605.18871","ref_index":66,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Distributional Energy-Based Models for Uncertainty-Aware Structured LLM Reasoning","primary_cat":"cs.LG","submitted_at":"2026-05-15T17:08:27+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A 149M-parameter distributional energy-based verifier with low-rank adapter ensemble reduces constraint violations in structured LLM reasoning and outperforms or matches much larger models on five benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.15775","ref_index":29,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Continual Learning of Domain-Invariant Representations","primary_cat":"cs.LG","submitted_at":"2026-05-15T09:31:10+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Introduces replay-based continual learning with sequential invariance alignment to learn domain-invariant representations, outperforming baselines on generalization to unseen domains across six datasets in vision, medicine, manufacturing, and ecology.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.13148","ref_index":54,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Understanding Generalization through Decision Pattern Shift","primary_cat":"cs.LG","submitted_at":"2026-05-13T08:14:59+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"DPS quantifies deviation of per-sample decision patterns from class averages and shows linear correlation with generalization gaps while unifying degradation scenarios into a continuous trajectory.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.12022","ref_index":12,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"SAGE: Scalable Automated Robustness Augmentation for LLM Knowledge Evaluation","primary_cat":"cs.CL","submitted_at":"2026-05-12T12:09:05+00:00","verdict":"UNVERDICTED","verdict_confidence":"UNKNOWN","novelty_score":6.0,"formal_verification":"none","one_line_summary":"SAGE trains a rubric-based verifier and an RL-optimized generator on seed human data to scalably augment LLM knowledge benchmarks, matching human-annotated quality on HellaSwag at lower cost and generalizing to MMLU.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.11161","ref_index":149,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Interpretability Can Be Actionable","primary_cat":"cs.LG","submitted_at":"2026-05-11T19:08:21+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Interpretability research should be judged by actionability—the degree to which its insights support concrete decisions and interventions—rather than explanatory power alone.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.11134","ref_index":6,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training","primary_cat":"cs.LG","submitted_at":"2026-05-11T18:41:12+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Characterizes spurious correlation mechanisms in preference optimization via mean spurious bias and causal-spurious correlation leakage, demonstrates irreducible vulnerability to distribution shift, and introduces tie training as selective mitigation with validation on log-linear models and empirica","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"characterization of model quality under the local regime. Margin definition.The expected margin measures the average score gap between preferred and dispreferred re- sponses: mD(θ) :=E (x,yw,yl)∼D[β ˜θ⊤∆ϕ]. The margin decomposes into causal and spurious components: mD(θ) =β ˜θ⊤ c µ(D) c| {z } =:mcausal D (θ) +β ˜θ⊤ s µ(D) s| {z } =:mspurious D (θ) .(6) We now show that margin differences provide a first-order approximation to objective differences. Proposition 5.1(First-Order Margin Approximation).Un- der the local regime (Assumption 3.2), the DPO objective ˜J(D)(θ) :=E D[logσ(β ˜θ⊤∆ϕ)]satisfies ˜J(Q)(θ)− ˜J(P) (θ) = 1 2 mQ(θ)−m P (θ) \u0001 +O(β 2∥˜θ∥2). The proof follows from Taylor expansion of the log-sigmoid;"},{"citing_arxiv_id":"2605.10546","ref_index":36,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Higher Resolution, Better Generalization: Unlocking Visual Scaling in Deep Reinforcement Learning","primary_cat":"cs.LG","submitted_at":"2026-05-11T13:23:58+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Higher-resolution observations with global-average-pooling encoders improve RL performance and generalization by enabling more localized visual attention, yielding up to 28% gains over standard Impala encoders.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.08519","ref_index":148,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"SeBA: Semi-supervised few-shot learning via Separated-at-Birth Alignment for tabular data","primary_cat":"cs.LG","submitted_at":"2026-05-08T22:03:31+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"SeBA is a joint-embedding framework that separates tabular data into two complementary views and aligns one view's representations to the nearest-neighbor structure of the other, improving feature-label relationships and achieving SOTA results in most benchmarks without relying on augmentations.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.06524","ref_index":48,"ref_count":2,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Process Matters more than Output for Distinguishing Humans from Machines","primary_cat":"cs.AI","submitted_at":"2026-05-07T16:30:35+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A new battery of 30 cognitive tasks demonstrates that process-level behavioral features distinguish humans from frontier AI agents better than performance metrics (mean AUC 0.88), with process-specific fine-tuning improving mimicry but limited cross-task transfer.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.02658","ref_index":5,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Deciphering Shortcut Learning from an Evolutionary Game Theory Perspective","primary_cat":"cs.AI","submitted_at":"2026-05-04T14:39:10+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Evolutionary game theory shows gradient descent and stochastic gradient descent drive neural networks to distinct stable states favoring shortcut or core subnetworks, with data and optimization noise shaping shortcut bias formation.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}