Chaos Is a LADDER: Domain Generalization Beyond Invariance via Reweighting

Domain generalization (DG) aims to learn from multiple source domains and generalize to unseen target domains. Most DG methods pursue invariance: they seek a causal representation whose prediction rule is invariant across domains. This principle is effective when the causal mechanism is stable, but becomes restrictive when the domain itself modulates how causal content maps to the response. In this case, directly feeding domain style into the predictor can create misleading shortcuts, since style does not by itself cause the response. Yet the apparent chaos of multiple styles can become a ladder: style can locate the unseen target domain among source domains and guide which domain-dependent prediction rules should be trusted. We propose \emph{Latent Adaptive Domain Disentanglement and Environment Reweighting} (LADDER), a fixed-model DG pipeline that learns causal/style representations, freezes the encoders, fits source-specific classifiers, and uses an unlabeled target-domain covariate set only at inference to compute weights over these fixed classifiers, with no target labels or model-state updates. We establish theoretical guarantees for source reweighting and validate LADDER on simulations, FMoW, and a location-grouped iWildCam protocol, with gains in overall and group-averaged accuracy.

GeoSelect: Spatial-Program Execution for Training-Free Referring Remote Sensing Image Segmentation

Referring remote sensing image segmentation isolates the object named by a natural-language expression in an aerial image. Existing training-free methods resolve the expression through implicit vision-language activations or region-text similarity, which gives weak control over the spatial, comparative, and ordinal relations that dominate aerial referring: they cannot represent constructions such as the largest ship or the second court from the left. We propose GeoSelect, a training-free pipeline that reframes referring as the execution of a typed spatial program. A frozen, text-only language model synthesises the expression into a small domain-specific language, a well-formedness checker accepts the program, and a deterministic executor runs it. The central abstraction is a single scored candidate set type under which every operator composes: continuous geometric fields realise position and proximity as dense pixel-level maps, while discrete set and order operators add the extremum, ordinal, counted-union, and relational constructions that fields alone cannot express. Because execution is explicit, every intermediate program, field, and ranking is inspectable, and a reliability ladder degrades any failing program to a field-only special case, so every expression returns an answer. GeoSelect attains 58.86 mIoU on RRSIS-D test and 55.27 mIoU on RISBench test, more than twice the best prior training-free method on RRSIS-D, with no referring supervision and on a single GPU. A controlled comparison with candidates and segmenter fixed attributes the gain to explicit execution, not the backbone; an oracle decomposition localises the residual gap to detection recall on RRSIS-D and selection on RISBench, and an exposure audit confirms robustness to pretraining leakage. Code will be released upon acceptance at the project page https://avalon-s.github.io/GeoSelect/.

The Scissors Effect: When Resize-Based Input Diversity Helps or Hurts Transfer Attacks

Input Diversity (DI), which applies random resizing and padding at each attack iteration, is a near-default ingredient of transfer-based adversarial attacks, widely assumed to improve transferability. We show this assumption is regime-dependent and, for robustly trained surrogates, often reversed. Varying only the surrogate, increasing the DI probability raises transfer success for standard surrogates but lowers it for robust ones: the two response curves separate like a pair of scissors, a pattern we call the Scissors Effect. The effect is strong and consistent on ImageNet, where blind DI costs the robust source 10.3% attack success on average across CNN, ViT, Swin, and ConvNeXt targets and across ten attacks spanning 2018-2024; it is smaller on CIFAR-10 unless DI is made aggressive. A controlled robustness-strength sweep that varies only the training budget shows the harm is graded rather than binary, crossing from beneficial to harmful already in the little-robustness regime. We trace it to gradient geometry: a resize/translation decomposition attributes roughly 67% of the harm to resize, and a direct source-target gradient-alignment measurement confirms the same resize operation improves alignment for standard surrogates but degrades it for robust ones. We summarize the regime with Local Gradient Consistency (LGC), a single input-space probe that separates the two surrogate types, and prove a bias-variance crossover theorem isolating where DI helps from where its resize bias dominates. A training-free rule (CG-DI) that disables diversity when LGC is high avoids the loss on robust surrogates while keeping DI's benefit on standard ones, positioning the Scissors Effect as a DI-specific manifestation of the broader robustness-transferability trade-off.

TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning

Reinforcement learning with verifiable rewards (RLVR) is a promising approach for enhancing reasoning and agentic behavior in large language models. However, rollout-intensive policy optimization is often limited by insufficient reward contrast, arising when overly simple or complex prompts generate low-variance feedback and when outcome-only rewards assign the same terminal assessment to every decision in a multi-turn rollout. Past efforts have focused on allocating available rollout resources to promising prompts, yet they only leverage sample informativeness at the prompt level and neglect variation in prefix-level informativeness across turns within the same rollout. This work targets multi-turn agentic RL by modeling each ReAct-style thought-action-observation turn as a semantically distinct node, allowing budget allocation to extend from prompt roots to turn-level prefixes with further continuations, which naturally forms tree-structured rollouts. We introduce Tree Rollout Allocation for Contrastive Exploration (TRACE), a unified rollout allocation framework that enhances reward contrast within a fixed sampling budget. Technically, TRACE allocates rollout budget to both prompt roots and intermediate prefixes that are most likely to yield mixed terminal rewards. A shared generalizable predictor estimates conditional success probability at these anchors from prefix histories to guide this allocation. The resulting adaptive tree structure enriches outcome-only feedback and amplifies the policy-update signal. Empirically, TRACE achieves competitive performance and efficiency gains on typical agentic benchmarks, e.g., improving Qwen3-14B Multi-Hop QA average accuracy by 2.8 points over competitive baselines at equal sampling cost.

Task Structure Reverses Layerwise State Encoding in Sequence Models

Mechanistic studies of sequence models often treat layerwise state encodings as architectural traits: recurrent models concentrate readable state, attention-based models distribute it. We find that the same architecture reverses this profile when the task changes. Across Transformers, Mamba, Mamba-2, LSTMs, and GRUs, Parity is concentrated late in Mamba and the recurrent baselines and built gradually by Transformer; on bounded-depth Dyck-k the pattern flips. The same flip appears in fine-tuned Mamba-130M and Pythia-160M, and the Pythia Dyck bottleneck persists at 410M. Two explanations are conflated in the literature: algebraic structure (commutativity) versus computational structure (prefix update vs. stack). To separate them we add a third task: non-commutative S_3 permutation composition. S_3 groups with Parity, not Dyck, on layerwise probing across all five architectures and on Mamba-specific Conv1D attribution, so the grouping tracks computational structure rather than commutativity. Causal interventions show that, in the 4-layer formal models, linearly readable directions are often functionally necessary and can remain important at out-of-distribution lengths on Parity and Dyck. At pretrained scale the picture splits. Fine-tuned Pythia Dyck has a strong middle-layer bottleneck (L6-L7 ablation drops accuracy by roughly 81% at 160M; broader L4-L18 plateau at 410M), far weaker at the best-probe layer. Pretrained Mamba shows the complementary failure mode: its final layer is highly readable, no single probe direction breaks the task on Parity, Dyck, or S_3, yet mid-position activation patching there recovers about 97-98% of the clean-corrupted logit gap. Probing localizes where state is linearly available, not always where the computation is bottlenecked. Mechanistic signatures are properties of architecture and task together.

Detection vs. Execution: Single-Bucket Probes Miss Half the Mamba-2 State Sink

Mechanistic interpretability often assumes that probes identifying a representational signature also identify the circuit executing the corresponding computation. We show that this assumption can fail systematically in Mamba-2. Studying the state sink (disproportionate Delta-gate activation on boundary tokens, analogous to the attention sink), we find that single-bucket probes recover only a small execution layer while missing a much larger detection layer with the same representational signature. In Mamba-2, the state sink decomposes into two functional head sets. Single-bucket BOS-specialist heads (about 5% of heads at 2.7B) causally support both BOS-context and newline-target predictions across model scales and corpora. Dual heads (27-35% of heads, recovered by multi-class aggregation of the same probe) show stronger BOS-newline representational similarity but substantially weaker causal effects under ablation. Representational similarity does not imply functional equivalence. This distinction matters for downstream behaviour: ablating BOS-specialist heads collapses RULER NIAH retrieval accuracy from 1.00 to 0.00 at 1024 context length in both Mamba-1 2.8B and Mamba-2 2.7B, while size-matched complements preserve baseline performance. A random channel-bucketing control rules out substrate granularity alone, implicating Mamba-2's head-shared Delta projection. Probe-derived specialty can identify execution circuits; at coarse granularity the same probe also recovers detection circuits, and separating them requires class-conditional ablation rather than class-conditional cosine.

PInVerify: An Offline Embodied Benchmark for Active Instance Verification

Embodied agents have made strong progress in navigating to target objects, but reaching the goal vicinity does not guarantee that the agent has found the correct instance: subtle attribute differences (e.g., "white floral" vs. "white striped") often require close-range, multi-view inspection. We address this gap with Active Instance Verification (AIV), a task in which an agent actively selects viewpoints around a candidate object to decide whether it matches a fine-grained natural-language description. We formalize AIV as a finite-horizon decision process and introduce PInVerify, an offline embodied benchmark for AIV: 3,000 evaluation episodes across 18 object categories, delivered as multi-view captures with a 6-sector navigation topology that exposes trap views (navigable but uninformative) and unreachable sectors. As reference baselines we build a training-free pipeline and a LoRA-fine-tuned end-to-end agent around open-source multimodal large language models (MLLMs) at on-device scale ($\leq$8B parameters), with attribute decomposition, a visibility-weighted multi-view tracker, and three next-best-view (NBV) strategies. In our evaluation across Qwen3-VL (4B/8B), SenseNova-SI-1.2-InternVL3-8B, CLIP, and SigLIP2, the best MLLM-based baseline exceeds the best embedding baseline by 4.9 pp; GT-box ablations show a +3.1 pp detection gap; and we do not observe reliable gains from active viewpoint selection within the tested NBV strategies. A LoRA-fine-tuned agent (SFT+GSPO) reaches 85.6%. PInVerify aims to support further work on active, fine-grained semantic verification in embodied AI. Code: https://github.com/Avalon-S/PInVerify.

Listwise Policy Optimization: Group-based RLVR as Target-Projection on the LLM Response Simplex

Reinforcement learning with verifiable rewards (RLVR) has become a standard approach for large language models (LLMs) post-training to incentivize reasoning capacity. Among existing recipes, group-based policy gradient is prevalent, which samples a group of responses per prompt and updates the policy via group-relative advantage signals. This work reveals that these optimization strategies share a common geometric structure: each implicitly defines a target distribution on the response simplex and projects toward it via first-order approximation. Building on this insight, we propose Listwise Policy Optimization (LPO) to explicitly conduct the target-projection, which demystifies the implicit target by restricting the proximal RL objective to the response simplex, and then projects the policy via exact divergence minimization. This framework provides (i) monotonic improvement on the listwise objective with bounded, zero-sum, and self-correcting projection gradients, and (ii) flexibility in divergence selection with distinct structural properties through the decoupled projection step. On diverse reasoning tasks and LLM backbones, LPO consistently improves training performance over typical policy gradient baselines under matched targets, while intrinsically preserving optimization stability and response diversity.

Closing Reasoning Gaps in Clinical Agents with Differential Reasoning Learning

Clinical decision support requires not only correct answers but also clinically valid reasoning. We propose Differential Reasoning Learning (DRL), a framework that improves clinical agents by learning from reasoning discrepancies. From reference reasoning rationales (e.g., physician-authored clinical rationale, clinical guidelines, or outputs from more capable models) and the agent's free-form chain-of-thought (CoT), DRL extracts reasoning graphs as directed acyclic graphs (DAGs) and performs a clinically weighted graph edit distance (GED)-based discrepancy analysis. An LLM-as-a-judge aligns semantically equivalent nodes and diagnoses discrepancies between graphs. These graph-level discrepancy diagnostics are converted into natural-language instructions and stored in a Differential Reasoning Knowledge Base (DR-KB). At inference, we retrieve top-$k$ instructions via Retrieval-Augmented Generation (RAG) to augment the agent prompt and patch likely logic gaps. Evaluation on open medical question answering (QA) benchmarks and a Return Visit Admissions (RVA) prediction task from internal clinical data demonstrates gains over baselines, improving both final-answer accuracy and reasoning fidelity. Ablation studies confirm gains from infusing reference reasoning rationales and the top-$k$ retrieval strategy. Clinicians' review of the output provides further assurance of the approach. Together, results suggest that DRL supports more reliable clinical decision-making in complex reasoning scenarios and offers a practical mechanism for deployment under limited token budgets.

Small Generalizable Prompt Predictive Models Can Steer Efficient RL Post-Training of Large Reasoning Models

Reinforcement learning enhances the reasoning capabilities of large language models but often involves high computational costs due to rollout-intensive optimization. Online prompt selection presents a plausible solution by prioritizing informative prompts to improve training efficiency. However, current methods either depend on costly, exact evaluations or construct prompt-specific predictive models lacking generalization across prompts. This study introduces Generalizable Predictive Prompt Selection (GPS), which performs Bayesian inference towards prompt difficulty using a lightweight generative model trained on the shared optimization history. Intermediate-difficulty prioritization and history-anchored diversity are incorporated into the batch acquisition principle to select informative prompt batches. The small predictive model also generalizes at test-time for efficient computational allocation. Experiments across varied reasoning benchmarks indicate GPS's substantial improvements in training efficiency, final performance, and test-time efficiency over superior baseline methods.