LLM 理论进展

A graph foundation model generalizes across graph domains by mapping every input into one shared representation before any task reasoning. We call this map the alignment layer, the component that separates a graph foundation model from a graph neural network, and we show it is a distinct attack surface that prior work has not studied. We attack it at inference time, with no access to training, on six public models spanning spectral tokenizers, text embedding spaces, and a discrete codebook. A directed representation-space perturbation collapses every model, but at a budget comparable to the representation norm a plain graph network also needs, with one exception: OpenGraph, whose spectral tokenizer collapses at a fifth of that budget, an alignment-specific fragility a plain network does not share and which a same-representation control traces to the tokenizer rather than the decoder. A realizable input-space attack that edits edges, features, or text removes at least half the correct predictions on three of the six models at peak. How much of this fragility an input-access attacker realizes tracks how directly the decoder reads the representation, and not the clean accuracy a task leaves; we measure this carrier gain structurally from the decoder's local Lipschitz sensitivity, and report clean-accuracy headroom as a within-model ordering heuristic that does not survive on realizable attacks.

Large language models (LLMs) are increasingly used for a range of software, hardware and human-centered security tasks. Consequently, LLM performance on security tasks is an active area of measurement and research, often with a focus on identifying areas in which LLM security ``knowledge'' may be insufficient. Popular strategies for identifying LLM security knowledge gaps include building corpora of challenge questions or task benchmarks, strategies that require substantial manual work and security expertise to design and execute. We introduce a partially-automated method for assessing LLM knowledge of a security area.The method uses authoritative information from Consumer Protection Agencies (CPAs) to identify instability in LLM responses that can be indicative of knowledge gaps. We demonstrate the method for 2 security topics, identity theft and impostor scams, and 5 LLMs in 2 leading LLM families, Gemini and GPT, using publicly available information about identity theft and impostor scams from 6 CPAs.The method distinguishes between models that have and don't have sufficient knowledge to accurately identify the security topics in text narratives.

Bug resolution is a cross-artifact process: natural-language reports must expose actionable behavioral cues that can be reproduced by tests and corrected by patches. Yet the extent to which these signals are preserved across artifacts remains largely unquantified. We introduce Desc2Fix, a framework for measuring semantic alignment between bug reports, triggering tests, and developer-written fixes. Alignment is operationalized through structured behavioral anchors (e.g., reproduction steps, API/exception cues, expected vs. actual behavior), deterministic similarity metrics (ROUGE, SBERT, CodeBERT, OpenAI embeddings), and LLM-based judgments grounded in coverage, correctness, and specificity. We analyze 2,857 report-test-patch triplets from Defects4J and SWT-Bench using GPT-4o and DeepSeek-Chat. LLMs reliably extract structured signals (up to 90% completeness) and exhibit strong cross-model consistency, yielding a stable semantic input contract for downstream reasoning. However, alignment is highly representation-sensitive: lexical similarity alone is insufficient; full diffs provide the most stable basis for judging report-patch correspondence; and structured summaries trade surface similarity for predicate-level salience. Across more than 182,000 LLM alignment ratings, both models show systematic optimism relative to humans (1-2 points on 5-point scales) and only modest rank agreement, motivating bias-aware evaluation. Our results demonstrate that behavioral alignment is measurable but not reducible to similarity, and that structured anchors combined with embedding-based proxies provide reproducible signals for ranking and filtering tests and candidate patches. By transforming alignment into a controllable engineering signal, Desc2Fix enables more reliable test generation, semantics-aware fault localization, principled patch ranking, and improved bug report authoring.

Diffusion Transformers achieve strong image generation performance, but most operate in compressed latent spaces. Pixel-space diffusion avoids this information loss, yet existing approaches map each raw image patch to a single token, forcing one representation to handle both global communication and fine-grained details. We address this issue by proposing a new architecture, DuSPiT, a Dual-branch SubPatch Pixel Transformer. This model separates global structural reasoning from local appearance modeling. DuSPiT uses a compact base branch for efficient global reasoning and a parallel, high-capacity pixel branch, organized into subpatch groups, to preserve detailed appearance, with the two branches interacting through cross-attention. Our results show that DuSPiT generates images with richer details and stronger fine-grained structures, while also achieving a better quality--efficiency trade-off than prior pixel-space diffusion transformers.

We present an exploratory study on using Large Language Models (LLMs) to generate Alloy formal specifications from both requirements documentation and production source code, and to derive executable test cases from those specifications. We evaluate on two real open-source Python libraries: Flipper, a feature flag management system, and Cerberus, a data validation library. In both cases, the LLM produced workable Alloy specifications and executable tests without any manual correction. For Flipper, our pipeline uncovered a genuine bug that the existing test suite had missed: the library silently accepts duplicate flag names, directly contradicting its documented uniqueness requirement. A direct LLM baseline--generating tests from the same README but skipping the Alloy step--achieved 68% branch coverage yet failed to catch this bug across all three independent runs. This suggests that introducing a formal intermediate representation can surface constraint-level defects that coverage-oriented generation may miss. For Cerberus, the code-derived specification captured an implicit abstraction over sized types that the documentation-derived spec omitted, producing two additional tests. Across both libraries, code-based specifications showed lower variance in test generation (mean SD = 2.15) than documentation-based ones (mean SD = 5.0), though whether this generalises remains an open question. Index Terms--formal specifications, Alloy, large language models, automated testing, specification drift, software validation.

Discourse relations provide document structure, critical to language understanding and enabling language model performance and ethicality. In this work, we investigate how instruction-tuned Transformer models (LLaMA and Mistral) encode discourse relations in English, with a particular focus on the contrasting relations of causation and antithesis. Framing the task as a next-token prediction task and applying a suite of interpretability techniques to test model internals, our findings show that certain early layers make predictive decisions at mid-sequence tokens, while some mid-level layers finalize their decisions closer to the last token. Most of the remaining layers primarily propagate earlier decisions rather than actively influencing them. Additionally, we observe that some layers exhibit a preference for one answer over alternatives, suggesting asymmetric representation of discourse-based reasoning.\footnote{Our code is available at https://github.com/abhidipbhattacharyya/causation_vs_antithesis}

Pragmatic language use requires reasoning about alternatives: the alternative expressions a speaker might have chosen, or the alternative interpretations a listener might entertain. Formal and computational models of pragmatics must therefore specify the sets of alternatives that interlocutors reason over, which is often done through manual specification. Here we propose a framework, ScAffolded Generative models for Explanation (SAGE), that combines the explanatory transparency of cognitive models with the generative flexibility of language models (LMs). SAGE decomposes a pragmatic process into three kinds of modules: proposers, which use LMs to generate an open-ended space of candidate alternatives; evaluators, which assess those alternatives (e.g., their semantics, complexity, or typicality); and selectors, which implement the rule-based computational steps of a cognitively motivated task analysis. We assess SAGE in three case studies spanning pragmatic generation and interpretation-referential expression generation, manner (M-)implicatures, and Gricean conversational implicatures. SAGE models are evaluated critically using established methods from computational cognitive modeling, including ablations, baseline comparisons, and quantitative fit to human data. Across studies, SAGE models achieved high accuracy and often outperformed baselines, but component-level analyses reveal an asymmetry: LM proposers reliably generated alternatives well-suited to pragmatic modeling, whereas LM evaluators are better at providing intuitive judgements rather than judgements of theoretical or formal measures. We discuss the promise and the limitations of neuro-symbolic models as candidate explanatory accounts of human pragmatic language use.

Large language models (LLMs) largely rely on Transformers, where self-attention provides global token interaction but does not explicitly encode the locality of natural language. We study whether lightweight depthwise convolutions can supply this local inductive bias without materially increasing model size. Our macro-level ablation compares convolution at 17 locations in a Qwen3 Transformer block and finds the best results when convolution is applied to the projected queries, keys, and values before attention. A subsequent micro-level study favors a residual depthwise convolution with kernel size $k=3$, without additional normalization or activation. Across Qwen3 models and several pre-training data budgets, this design improves the average accuracy on seven downstream benchmarks while adding less than $0.01%$ parameters. A representation-level case study further suggests that the convolution makes repeated token IDs more sensitive to their immediate context. These results support depthwise convolution as a lightweight complement to self-attention for modeling short-range token interactions.

Neural simulation-based inference enables parameter estimation for complex models, but typically requires the user to specify a simulator encoding a fixed model structure. We present a framework for joint model selection and parameter estimation that combines large language models for program synthesis with neural simulation-based inference. Given a natural language description of the system and data under investigation, an LLM proposes candidate simulator programs which are iteratively refined via feedback-driven mutation and evaluated using neural density estimation. The approach enables simulation-based inference over a pool of models, not just parameters within a fixed model. On benchmarks spanning deterministic dynamics, stochastic epidemic models, and dark matter substructure inference from gravitational-lensing images, the method identifies plausible model families from open-ended prompts, with accuracy that reflects the information content of the data and identifiability of candidate models.

Cyber-physical systems built on deterministic edge inference, such as on-vehicle flood detection for agricultural fields, produce structured decision logs that must be interpreted differently by heterogeneous stakeholders. Pairing such systems with large language models (LLMs) to generate stakeholder-specific reports introduces a tension: the generative layer is non-deterministic, while the edge plane must remain replayable and auditable. We propose an architectural pattern resting on two invariants: unidirectional consumption, in which the generative layer is a strict read-only consumer of the deterministic plane and never writes back, and persona-as-configuration, in which stakeholder adaptation is a versioned prompt-template artifact rather than runtime improvisation. We instantiate the pattern as a context-aware dashboard layer over the JSON decision logs of a previously published edge-based standing-water detection system, and analyse how the integration boundary admits standard generative-reliability mitigations as configuration- or middleware-level extension points. A structured expert review rated the pattern favourably across five ISO/IEC 25010-aligned quality dimensions, with strongest agreement on separation of concerns. End-user evaluation with agricultural stakeholders is planned for future work.

Enterprises will not deploy AI agents they cannot trust, and the most-cited reason for distrust is hallucination: confident, fluent output that is simply not true. The common response is to wait for a model that does not hallucinate. We argue that this is the wrong target. Large language models are, by construction, capable of generating unsupported text, and no amount of scale removes the possibility; a faithfulness judge bolted onto a raw model catches some errors but still ships others, and even well-curated retrieval pipelines have been shown to fabricate citations. We reframe the goal: "zero hallucination" is not a property a model possesses but a property a system enforces. We present HALO (Hallucination-Aware Layered Oversight), an assurance architecture which treats hallucination as a containable failure mode rather than an eliminable one. HALO composes six layers of defense: grounded generation over retrieved, approved content; constrained, deterministic execution that bounds where the model can err; multi-signal verification that scores every output for groundedness and hallucination using both an LLM judge and evidence-based checks against the source text; calibrated abstention, so the system declines rather than guesses when grounding is insufficient; total traceability of every retrieval, tool call, and generation; and continuous oversight that detects drift, alerts on threshold breaches, and closes the loop by regenerating and statistically validating improved agents. We detail each layer, give particular attention to evidence-based confidence (which verifies extractions against the source document rather than trusting the model's self-reported certainty), and illustrate the architecture on a regulated claims-extraction workload

Does adding a reasoning step make a Vision-Language-Action (VLA) model more robust to perturbation? Intuitively, a policy that reasons before acting should absorb a perturbed input better than one that maps observations directly to actions. We test this premise head-on across three models that span the reasoning spectrum (no reasoning, a text chain-of-thought, and a latent iterative loop), perturbing each at the vision, reasoning, and action stages on LIBERO and SimplerEnv. Two questions organize the study: does the reasoning design shift robustness, and can the reasoning be read back at runtime as a safety signal? We find that the latent-iterative model is by far the least robust: under both stochastic noise and white-box perturbation its task success collapses, while the other two hold. This fragility is structural rather than cumulative: varying the reasoning depth at inference barely moves it. Reasoning outputs can in principle be monitored, but the monitors fail under fair tests. A plan--action consistency probe that looks near-perfect under naive evaluation falls to chance under adaptive attack. Under matched-FPR calibration, fusing it with an action-anomaly probe never lifts defended success above undefended. Scoped to these output-level behavioral probes under white-box vision-stage attack, this ceiling is a precondition that any viable defense must first satisfy.

Representation is a central concept in modern machine learning, where it usually refers to internal encodings that support learning and generalization. As models scale and their capabilities become increasingly human-level, this representational language sometimes shifts from an engineering context into the more philosophically loaded domain of mental representation. We argue that this is the case for recent claims about the convergence of representational properties across different AI models. In particular, we assess the arguments developed in The Platonic Representation Hypothesis, according to which this convergence is driven by a unified structure of reality. We examine this claim by introducing arguments and ideas from debates about mental representation in the philosophy of mind. We argue that these philosophical resources can clarify what is at stake in such claims, explain why alignment evidence alone is insufficient for strong metaphysical conclusions, and suggest directions for future research.

LLM-based autonomous agents require external memory to overcome their statelessness and limited context window for long-term interaction and dynamic knowledge reasoning. However, existing memory retrieval methods often lack adaptability and sample efficiency, and struggle to retrieve the right mixture of memories from heterogeneous stores. We propose Exploratory-Assimilating Reflection (EAR), a framework for high initial retrieval performance and sample-efficient adaptation. EAR combines two mechanisms: Exploratory Reflection, which performs iterative search to bootstrap retrieval and collect useful experiences for each query, and Assimilating Reflection, which replays these experiences from an Experience Buffer to refine a global reranker more efficiently than methods relying only on immediate rewards. Experiments show that EAR improves retrieval by up to 17.9% over the baseline retriever on two long-term dialogue benchmarks. We also show that EAR is highly sample-efficient and robust to noisy feedback.

Highly nonlinear chaotic dynamical systems remain difficult to model due to fundamental trade-offs between complexity, expressivity, and data efficiency. Modern machine learning methods achieve strong predictive performance but often rely on a-priori system knowledge or curated data with limited interpretability. Koopman operator theory offers a promising direction via linear representation in an infinite-dimensional observable space. However, many data-driven Koopman methods seek globally valid operators for which useful finite-dimensional spectral embeddings remain difficult to identify under these constraints. To overcome associated limitations, we introduce Fuzzy Spectral Region Decomposition (fSRD), a fully automated learning framework for estimating finite Koopman representation via multiple operators. The proposed method realizes a data-adaptive framework for assembling locally invariant embeddings, termed Invariant Decomposition. fSRD achieves highly accurate linear reconstructions of nonlinear systems while learning finite-dimensional representations of their induced evolution operators, bridging interpretable operator-theoretic models with expressive data-driven sequence learning. These embeddings are adaptively constructed via a global fuzzy tree model, drawing inspiration from fuzzy neural architectures to learn the induced dynamics while prioritizing parsimonious solutions. Empirical results across canonical chaotic systems (e.g., Lorenz and Duffing) and high-dimensional real-world data demonstrate strong predictive accuracy, interpretability, and robust expressivity across data-rich and data-limited regimes, highlighting the method's generality.

Structure-based drug design (SBDD) leverages the 3D structure of protein targets, often complemented by other spatial constraints, to generate candidate binding molecules. While diffusion models have dominated as a leading paradigm for high-quality 3D molecule generation, LLM-based methods are rapidly emerging in molecular design and have shown competitive performance in pocket-conditioned molecular generation. However, their ability to reason about physics and 3D spatial environments is largely underexplored. In this work, we systematically analyze whether current general-purpose LLMs are capable of navigating complex 3D constraints compared to established baselines such as specialized diffusion models. We consider 3D ligand generation conditioned on protein pockets together with ligand- and interaction-derived spatial constraints, including anchor fragments, pharmacophore points, and mandatory pocket-ligand interactions. To enable this evaluation, we introduce 3D-Fit - a token-efficient benchmarking strategy for assessing LLM performance on multi-conditioned spatial molecule generation. Our findings reveal a clear pattern in LLM spatial capabilities: while they still lag behind state-of-the-art approaches, they are promising and can handle multiple spatial constraints simultaneously, enabling scaling to heterogeneous setups.

Vision-language-action (VLA) models have achieved impressive generalization in robotic manipulation, and recent memory-augmented VLAs have relaxed the Markovian assumption by conditioning on past images or language summaries. Vision-based memory approaches address this by conditioning on sampled past image frames, but they are computationally expensive and fundamentally limited when temporal events are visually ambiguous, e.g., pushing a button multiple times with small movements. We propose FM-VLA, a VLA model with force-based memory, enabling temporal context reasoning for non-Markovian, contact-rich manipulation. We encode force histories into compact force memory tokens with a variational autoencoder (VAE) pretrained with force time series reconstruction. By projecting force latent representations and short state history as additional conditioning tokens to the action expert module, we enable VLAs to leverage accumulated contact event history to guide manipulation. We evaluate FM-VLA on three memory-dependent tasks, including finding a hidden block, pressing a button, and wiping a dish for a specific number of times. Our lightweight force memory achieves over 80% success rate with minimal inference overhead, significantly outperforming baseline approaches. Project page: https://qft-333.github.io/FM-VLA-Page/

Industrial Video Anomaly Detection (IVAD) aims to identify anomalous objects and events in an industrial process, which is crucial for modern manufacturing and quality control systems. Existing VLM-based anomaly reasoning methods are capable of detecting open-ended anomalies in general domains. However, their performance declines in industrial settings characterized by intricate object transformations, strict physics, and procedural constraints. To tackle the complexity of such interaction-intensive detection, we introduce a training-free agentic framework for anomaly detection free of domain-specific knowledge, emphasizing object state evolution like humans inspectors. It is designed to track spatial-temporal dynamics and underlying transformations of detected objects over time, and then reason over the object-wise temporal state trajectories to identify abnormal objects in grounded frames. Our method overcomes limitations of prior approaches that rely on retraining on normal clips or injecting domain knowledge as context for test-time inference. Extensive experiments on three IVAD datasets demonstrate that our method outperforms frontier VLMs, agentic frameworks, and traditional VAD methods fine-tuned on the respective datasets, while providing interpretable reports over anomaly processes and types.

Structured pruning compresses large language models (LLMs) by removing whole computational units, such as attention heads and feed-forward (FFN) channel groups. Most training-free methods, however, rank these units independently, implicitly treating the loss from pruning a set as the sum of its individual losses. This view fails for Transformers, whose sublayers are coupled through a shared residual stream. Two individually weak units can thus be jointly indispensable, yet independent scoring is blind to such dependence and removes them together. We introduce CoCurve (Cross-Module Co-Pruning Curvature), a calibration-only, fine-tuning-free method that prunes attention and FFN units jointly. A second-order Taylor expansion of the token-level KL between the frozen model and its masked copy yields a single Fisher matrix whose diagonal is classical node saliency and whose off-diagonal entries are co-pruning curvature edges: the extra damage of removing two units together. Under a single-ablation additivity approximation this matrix reduces to a Gram product of single-unit ablation features, so the full M x M interaction is recovered from M forward passes, with no pairwise sweeps or gradients. Pruning then reduces to one budgeted quadratic program, solved in a single shot under a shared attention--FFN budget, with no labels, fine-tuning, or recovery.

We propose ARBITER, a novel LLM guardrail framework that introduces two key ideas: (i) dual-hypothesis reasoning, a reasoning method for LLM guardrails that explicitly considers both safe and unsafe interpretations of a prompt before making a safety decision, and (ii) multi-component supervised fine-tuning (MC-SFT), a structured training loss for reasoning-based guardrails that decomposes LLM outputs into logical components and weights them according to their importance. Existing reasoning-based guardrails often rely on expensive procedures, such as generating reasoning traces using larger or closed-source teacher models and applying full-parameter fine-tuning. In contrast, ARBITER uses a cost-effective self-generation strategy for reasoning traces and LoRA-based parameter-efficient fine-tuning while still achieving better performance than these expensive approaches. Additionally, ARBITER provides faithful evidence-phrase explanations for unsafe decisions, enabling a more transparent and interpretable guardrail method. Experiments on three safety moderation benchmarks show that ARBITER outperforms existing reasoning-based and non-reasoning guardrail baselines, with clear gains in out-of-domain evaluations.

Large Language Models (LLMs) have had a remarkable impact across many areas of machine learning. However, recent studies have shown that they struggle to reliably solve planning problems. At the same time, theoretical results have shown that transformers, the core architecture underlying modern LLMs, are Turing-complete. In this work, we investigate this apparent gap between the theoretical computational power of LLMs and their empirical planning performance. We propose Chain of Computation (COC), a computational architecture that places a transformer-based LM inside an iterative loop, leveraging its strength as a pattern-matching system. The COC uses a Structured Context Window (SCW) which provides a constant-sized context window with support for choosing which window is used at each planning step. Within this architecture, the LM is able to learn a planning policy, predicts the world model, and performs the arithmetic operations required during planning. We show that, when given an append-only SCW (resembling a Turing Machine tape), even relatively small LMs trained from scratch can learn planning policies and generalize from a small number of training instances within each planning domain, achieving success rates above 99.89% on BlocksWorld and the Pancake puzzle. Our analysis of failure cases in Tower of Hanoi (TOH) reveals that they arise from arithmetic operations or from encountering previously unseen tokens. We show that COC can solve TOH problem instances with up to 20 disks, requiring over 1 million actions, while requiring substantially less training data by either (1) planning with symbolical support for arithmetic or by (2) using a deterministic pushdown automaton (PDA) formulation for the SCW.

Large language models (LLMs) are increasingly adopted for network automation, yet their output quality and inference cost can vary substantially across LLM families. We present HuGLEN, a stepwise evaluation pipeline that uses an LLM-as-a-judge together with a small set of expert ratings to enable scalable and reproducible comparison of candidate LLMs, and to rank them using a quality efficiency score (QES). We demonstrate HuGLEN for translating outputs from an explainable artificial intelligence (XAI) model for the optical network quality of transmission (QoT) estimation task into operator-friendly explanations. Our results show that a medium-sized LLM (12B parameters) achieves the highest QES, indicating the best trade-off between explanation quality and efficiency. Overall, HuGLEN reduces the human-labeling burden while supporting consistent model selection for operator-facing automation tasks.

Not all training samples contribute equally to large language model fine-tuning. Selecting informative training samples can reduce the computational cost while preserving downstream performance. Many existing data selection methods rely on indirect heuristics, such as data quality, diversity or reasoning trace length. However, the effectiveness of these fixed criteria is task-dependent and difficult to generalize across diverse downstream tasks. Perplexity-based data selection provides a simple and model-aware solution to estimate the sample difficulty, but existing approaches typically score the entire training sequence and ignore the difference in learning objectives of language modeling and reasoning tasks. In this paper, we propose PPL-Factory, a simple and interpretable data selection framework that combines task-aware perplexity-based scores and data budget-aware selection criteria. Experiments on GSM8K demonstrate that PPL-Factory outperforms other state-of-the-art data selection methods using only $1%$ of the training set. With $10%$ of the data, PPL-Factory exceeds full-data fine-tuning accuracy by 0.9 on GSM8K and 4.8 on MATH. Overall, our results demonstrate that task-aware and budget-aware perplexity-based selection provides an effective and applicable approach for efficient fine-tuning.

Extended reasoning has become standard for frontier Large Language Models (LLMs), yet the trajectories these models produce remain largely uncontrollable. Existing methods for shaping how a model reasons are prompt based approaches and operate at the input level, offering no fine-grained control over the reasoning process itself. Related work analyzes and discovers latent transition dynamics in the reasoning traces from Large Language Models. Building on this, we statistically characterize these states, and show that failure trajectories get stuck in self-loops, exhausting the token budget without progress toward the final answer. To intervene on these failures, We propose SOPHIA: Steering Of reasoning Processes via Hidden-state Intervention and Activations. We treat each reasoning trace as a sequence of latent states rather than an unstructured texts, and investigate whether inference time interventions can provide fine-grained control over the self-looping reasoning process. We classify every prefix to a latent state, record step level transitions, and use them to construct a bank of steering vectors indexed by state pairs. At inference time, a controller infers the current state and, given a target state, retrieves the corresponding vector and can also detect self-loops online from the transition structure to prevent the model from sinking into a reasoning black hole. Through extensive experiments, our method reliably intervenes on self-loop failures, with steering vectors that generalize to different state pairs. End task accuracy and token efficiency indicate that fine-grained controllability results in better reasoning quality.

Multimodal large language models (MLLMs) still struggle with spatial reasoning that requires perspective transformation. In particular, they often rely on camera-centric cues rather than reasoning from the reference object's viewpoint, leading to systematic errors in non-camera reference settings. In this paper, we first analyze this failure mode and show that object orientation is a key factor underlying such camera-centric shortcut behavior. To address this issue, we propose OrientSAM, an orientation-aware spatial alignment framework for multimodal models. OrientSAM injects explicit orientation information into multimodal representations through orientation-aware tokens and Fourier-based angle encoding, and further adopts a curriculum learning strategy to progressively improve perspective-aware reasoning. In addition, we build a spatial data construction pipeline to generate orientation-aware spatial supervision from large-scale images. Experiments on Spatial-MM, ViewSpatial, and 3DSRBench show that OrientSAM consistently outperforms strong baselines, especially on non-camera-view, person-centric, and orientation-sensitive tasks. The results further demonstrate that explicit orientation modeling is important for mitigating camera-centric shortcut behavior and enabling more robust allocentric spatial reasoning in multimodal models.

When does a committed intermediate stage in an LLM reasoning pipeline earn its cost? Constrained Path Reasoning (CPR) pairs a source-aware path hypothesis with stage-level accounting. Search generates provisional states; trusted or validated invariants can constrain hard, while other proposals remain soft and revisable. CPR predicts that task-compatible commitments can factor transitions, concentrate candidate mass, induce regularity, and expose feedback when their gains exceed propagated error and execution cost. The formalism covers discrete commitments and continuous flows and measures effective branching, endpoint concentration, and cost per usable output. Across 1,180 generated QCQPs and 40 engineered degenerate polynomial instances (2,140 endpoints), residual triage recovers 63.0% of repair-all's additional feasible yield with 17.7% of its attempts. Fixed-LLM accounting (270 unique calls shared across nested arms) finds usable yield of 41.1% direct, 90.0% after formalization and deterministic execution, 20.0% after one-shot convexification, and 21.1% for the full path. In 120 paired-condition calls, a two-action rollback rule reaches 90% usable yield versus 36.7% for the feedback-conditioned selector. Two endpoint probes separate source from validation: a 72-output cross-trajectory transplant reduces entropy and acceptable mass; a 24-output same-call self-proposal pilot gives unchanged two-repeat collision entropy, 25.0% versus 8.3% usable yield, and 1/8 deterministically confirmed endpoint checks. Model-generated states supply hypotheses; trusted execution earns constraint strength.

Efficient large language model (LLM) inference on edge platforms is limited not only by model size, but also by shape-dependent performance differences across execution backends. Static backend assignment cannot exploit this variation, while independent per-operator selection can introduce costly device and framework switches. This paper presents a transition-aware backend dispatch approach for edge transformer inference. The approach combines current operator features with the previously selected backend to preserve beneficial shape-specific choices while avoiding unnecessary transitions. Ordered traces are collected from full-model inference runs of seven transformer models, and four common operator classes are benchmarked across PyTorch eager CPU, PyTorch eager CUDA, and ONNX Runtime CPU on an NVIDIA Jetson platform. The dispatch policies are evaluated through measurement-backed trace replay using observed operator costs and transition costs measured from actual backend switches. Supported operators are selected dynamically, while operators outside the dispatch scope retain a static assignment. Across 9,584 ordered operator instances and 278 exact shape groups, transition-aware dispatch reduces replayed latency, energy, and energy-delay product relative to the best static policy by 17.4%, 14.4%, and 28.5% on average, respectively. It also reduces switching relative to operator-local selection. Leave-one-model-out evaluation improves all three objectives for six of seven held-out models and improves energy for all seven. These results demonstrate that incorporating operator shape and backend-transition context can improve selective backend dispatch for edge transformer workloads.

Recent advances in Multimodal Large Language Models (MLLMs) have significantly improved remote sensing (RS) multimodal understanding. Language-conditioned segmentation is crucial for fine-grained target understanding in Unmanned Aerial Vehicle (UAV) videos. However, this task remains challenging due to the prevalence of small, visually ambiguous targets and dynamic aerial perspectives. In this paper, we propose SkyVLaM, a multimodal large language model for UAV video understanding. SkyVLaM constructs sparse tokens directly from patch-level video representations through a temporal basis perceiver, regularizes the sparse basis to encourage complementary temporal cues, and adaptively selects a temporally coherent dense segment for high-resolution inspection. The resulting sparse and dense tokens are jointly processed by a large language model for query-conditioned segmentation. We further build SkyVid, consisting of SkyVid-VGCG and SkyVid-RVOS for video grounded conversation generation and referring video object segmentation, respectively. SkyVid contains 101 videos, 33.6K frames, and 1.53M pixel-level object instances. Experiments show that SkyVLaM provides a more effective allocation of the visual token budget and improves language-conditioned video segmentation in UAV scenarios.

We study the problem of sequentially evaluating a new large language model (LLM) on a fixed question set using historical performance data from prior LLMs. Our goal is to construct a confidence sequence (CS) for the model's capability on this question set and to design active querying rules that shrink the CS width as quickly as possible. For CS construction, we invert a family of test supermartingales and focus on two representative approaches: a reverse information projection (RIPr)-based approach and a testing-by-betting-based approach. We first study these approaches under an oracle setting, and demonstrate the oracle optimality of the RIPr-based construction. We then propose a growth-oriented querying rule that aims to maximize the worst-case one-step expected log-increment over the endpoints of the current CS. In practice, we build these test supermartingales and the querying rule on predictions of question-level correctness learned from historical data. We then analyze the shrinkage behavior of the resulting CSs and identify two key factors that slow the shrinkage rate of CSs: accumulated prediction mismatch and the spikiness of the querying distribution. Finally, motivated by this analysis, we propose several mixture querying rules that combine growth-oriented querying, prediction refinement, and uniform exploration, trying to mitigate the effects that slow the shrinkage rate. We provide experiments comparing different querying rules for the RIPr-based and testing-by-betting-based CSs across several synthetic testing datasets. Interestingly, we observe that the simplest querying rule, uniform sampling, can sometimes outperform more adaptive querying rules for both methods.

LLM constraint reasoners are often evaluated near the random-SAT phase transition, confounding density and solver hardness. We test instance-level transfer while near-matching clause density. At aligned size bins, with near-matched density and matched maximum clause width, we compare proof-hard expander-Tseitin and proof-easy ladder-Tseitin formulas, pigeonhole anchors, and density-mismatched controls. Theory separates their resolution hardness; a solver-specific Glucose mean-conflict proxy differs by up to $51\times$, and five other solvers preserve the direction. Across three included models (243 instances each; a fourth is excluded for abstention), the near-matched-density accuracy gaps range from $-32$ to $+20$ points, with a pooled gap of $+1.7$ points ($p=0.74$) and a wrong-signed correctness-versus-conflict association ($r=+0.15$). A proof-preserving relabeling lowers accuracy in all five clusters for one model (mean $-93$ points) but not another, exposing model-surface sensitivity. In a preregistered extension, provider-reported completion-token spend does not consistently increase with the proxy after accounting for formula length and censoring. At 16k, the reasoning model spends more on proof-easy matched formulas and exhausts its budget on the solver-easiest UNSAT family; the 32k C1 gap is absent. These scoped dissociations concern verdict accuracy and observed token spend, not certificate solving, exact proof length, or allocation efficiency.