Hallucination in Large Language Models (LLMs), characterized by the generation of content inconsistent with contextual facts or logical constraints -- remains a persistent challenge for reliable deployment. In this work, we address this issue through a geometric framework rooted in the linear representation hypothesis. We propose that hallucinations manifest as orthogonal noise relative to the semantic manifold of the residual stream. Specifically, we hypothesize that while attention heads ideally propagate information congruent with the context subspace, hallucinations arise when specific heads introduce components orthogonal to this subspace, disrupting the coherence of the latent representation. Based on this formulation, we introduce Dynamic Contextual Orthogonalization (DCO), an inference-time intervention method. DCO utilizes the input residual stream as a dynamic context anchor to perform orthogonal decomposition on attention head outputs. To distinguish between context-aligned semantic updates and divergent noise, DCO employs a layer-wise Z-score suppression mechanism that selectively attenuates outlier orthogonal components based on statistical distributions. Evaluations on Llama-3-8B and 70B across benchmarks such as XSum, NQ-Swap, and IFEval demonstrate that DCO achieves superior contextual faithfulness compared to state-of-the-art intervention baselines. Furthermore, DCO maintains high performance on knowledge-intensive tasks like TriviaQA and TruthfulQA, effectively mitigating the trade-off between hallucination suppression and parametric knowledge retention often observed in existing methods. Our findings validate the geometric interpretation of hallucinations and establish DCO as a computationally efficient approach for enforcing manifold alignment.Our code is available at https://github.com/Harry-Miral/DCO

Recent developments in Large Language Models (LLMs) have showcased impressive reasoning capabilities, with Reinforcement Learning with Verifiable Rewards (RLVR) being a promising enhancement strategy. However, existing reward mechanisms are constrained to the outcome-level correctness and lack explicit signals to guide the model to consider diverse solutions. In contrast, human problem solving typically involves evaluating multiple potential approaches and selecting the most reliable solution, a cognitive process that current RLVR frameworks do not explicitly incentivize. Inspired by this, we propose Hint-Guided Diversified Policy Optimization (HDPO), allowing the model to first list all potential candidate solution outlines as hints and then select the most reliable one for further reasoning. HDPO comprises two stages of Cold Start for Structured Reasoning and Hint-Guided Diversified Reinforcement Learning to incentivize the model to generate diverse and reliable solutions following the ``propose-select-think'' trajectory. Experimental results show that HDPO effectively boosts LLM reasoning and enhances the diversity of candidate solutions as well as the LLM's ability to identify reliable solutions.

Reward transfer in Inverse Reinforcement Learning (IRL) is unreliable when policies must generalize to unseen combinations of environment dynamics and task goals. We propose Factorized Contrastive Abstractions for Transferable IRL (ConTraIRL), a framework that enables compositional reward transfer by learning decoupled latent representations of these two factors. ConTraIRL uses a dual-encoder architecture that maps observations into separate dynamics and goal latent spaces, trained with a dual contrastive objective. Temporal alignment encourages the dynamics encoder to learn goal-invariant structure, while the goal encoder captures dynamics-invariant features. This factorization supports reward inference under recombined dynamics-goal settings. Experiments on continuous control benchmarks demonstrate effective few-shot transfer to unseen dynamics-goal pairings, improving sample efficiency and reward recovery over transfer IRL baselines.

Mixture-of-Agents (MoA) systems improve reasoning accuracy by routing each query to multiple expert LLMs and aggregating their outputs. Efficiently executing this workload on limited GPU resources has bottlenecks. Skill-based routing creates skewed expert demand, and combining instruction-tuned LLMs with long-reasoning models results in extreme variability in generation lengths. Consequently, traditional scheduling strategies suffer from significant GPU idling and throughput collapse due to load imbalances. We present MOSAIC, a scheduling framework to accelerate MoA workloads. First, we formulate an Integer Linear Program (ILP) based scheduler that jointly optimizes expert placement and per-worker prompt assignment from offline-profiled costs, replicating reasoning experts across workers while pinning lightweight ones. Second, MOSAIC uses confidence-aware adaptive aggregation, leveraging inter-expert agreement to bypass the heavy final aggregator LLM for consensus queries. In our 4-GPU system, MOSAIC achieves up to 2.5x expert-stage, 4.23x aggregator-stage and 1.7~2.3x end-to-end speedups over the baseline scheduler, while matching accuracy within 0.1pp.

Despite rapid progress, multimodal large language models (MLLMs) still fail on tasks that humans solve effortlessly, such as navigating a grid maze from a screenshot or selecting the correct puzzle piece. Rather than retraining the model, we ask a complementary question: how much capability can be elicited from a frozen MLLM purely by improving the execution scaffold around it? We introduce MUSE, a multimodal unified structured execution harness that wraps any off-the-shelf MLLM with composable modules for task representation, visual processing, perception tool use, structured parsing, deterministic verification, and verifier-guided repair, without any model retraining. We evaluate MUSE across diverse benchmarks spanning visual spatial planning, visual perception, multimodal reasoning, and fine-grained visual discrimination, using multiple state-of-the-art MLLMs. MUSE delivers consistent gains over the bare model in all settings, with the largest jumps on challenging instances. Further analysis reveals that many MLLM failures arise from harness-level shortcomings rather than fundamental model deficits, and can be addressed through verifier-guided repair without touching the model. These findings highlight the agentic multimodal harness as a critical yet underexplored design dimension, offering an orthogonal avenue for improving MLLMs beyond model-centric optimization.

A latent world model built from an equivariant encoder $E$ and an equivariant predictor $f$ inherits a provable symmetry of its training loss: when the world's dynamics genuinely carries a group $G$ acting on latents by an orthogonal representation $ρ(g)$, the one-step prediction relMSE is exactly invariant across the whole group, so fitting the dynamics on a restricted slice of orientations mathematically determines it on the entire orbit (jǔ yī fǎn sān). We verify this end-to-end at laptop scale (CPU/MPS, fully seeded). [A] The symmetry survives a real Muon/AdamW + EMA + VICReg run -- composed encode-then-predict residual $\sim 10^{-6}$ after optimisation, not just at initialisation, and under any optimiser. [B] One-step error is flat to five digits across the group, while a same-hypothesis-class non-equivariant baseline fits the slice but breaks out-of-distribution (VN $\times 1.00$ vs baseline $\times 13.8$ in 2D, $\times 17.2$ in 3D, $\times 157$ over the full $\mathrm{SE}(3)$ ladder), with the equivariant model $4.5$-$7.4\times$ smaller. [C] The same isometry argument lifts to closed loop: under a matching equivariant planner the control trajectory at orientation $g$ is exactly $ρ(g)$ applied to the seen one, so closed-loop error is invariant across the group -- float-floor-exact in 2D/$\mathrm{SO}(2)$ on real PushT and statistically flat in 3D/$\mathrm{SE}(3)$ (disjoint 95% CIs). We stress-test the prior against Sutton's Bitter Lesson: augmentation, brute-force scale, and soft-equivariance each close at most the across-group task metric, never the float-floor exactness. Because equivariance is closed under composition, the $H$-fold rollout stays flat ($\times 1.00$, $\le 2\times 10^{-7}$) at every horizon, while the baseline's residual compounds with $H$. Out of scope: task-success sweeps, planner-free invariance, and scaling.

Automated cough analysis offers a path to low-cost respiratory screening, but most existing work stops at binary COVID-19 detection. A practical tool needs to tell apart several respiratory conditions from one cough recording on a consumer smartphone. We present CoughSense, a system that sorts cough recordings into five classes. These are healthy, COVID-19, asthma or respiratory condition, bronchitis, and pneumonia. We aggregated 18,301 recordings from four public datasets (Coswara, CoughVID, Virufy, and the West China Hospital Pediatric Cough Dataset) and used the OpenAI Whisper encoder as a pretrained backbone for cough disease classification. The main contribution is active-frame QKV attention pooling, which restricts attention to the first 200 of 1500 encoder tokens. This avoids the silence-dilution problem that arises because a 3-second cough fills only 150 tokens of Whisper's 30-second input window. Other training parts handle the 19 to 1 class imbalance and the four-dataset domain shift. These include WeightedRandomSampler, SpecAugment, Balanced Mixup with forced minority pairing, a supervised contrastive auxiliary loss, FiLM symptom conditioning, and gradient-reversal domain adaptation. A dual-encoder model fuses Whisper with the OPERA-CT respiratory foundation model through cross-attention. CoughSense (Whisper-tiny, 8.6M parameters) reached 82.3 percent balanced accuracy on five-fold cross-validation (macro-F1 of 0.817, AUC of 0.941). It beat an ImageNet-pretrained EfficientNet-B2 by 11.1 points and a ViT trained from scratch by 29.6 points. All five classes passed 74 percent recall and four of five passed 80 percent. The dual-encoder model reached 85.4 percent balanced accuracy. Active-frame pooling is the largest single contributor across all ablation components at 5.1 points, which should help any short-audio task using Whisper as a backbone.

Inertial odometry (IO) using only Inertial Measurement Units (IMUs) provides a lightweight solution for human motion tracking in augmented reality (AR) and wearable devices. Recent learning-based IO methods have improved the generalizability of inertial localization through large-scale pretraining on human motion datasets. However, these approaches remain prone to drift and noise because they do not explicitly capture human motion dynamics, especially on daily activity datasets such as Nymeria. In this work, we propose to ground inertial odometry in human kinematics through a learned IMU-inferred pose prior, which promotes physically consistent motion constraints. We integrate this pose prior into existing IO architectures and reduce positional drift by up to 36% on the challenging Nymeria dataset, which is 5x larger than datasets used in prior work. We further improve long-term performance with a sensor-fusion framework that incorporates auxiliary signals from lightweight sensors already available on commercial AR glasses, including magnetometers, barometers, and secondary IMUs. With this fusion strategy, positional drift is reduced by up to 42%, improving robustness and generalization across diverse motion conditions. Together, our results introduce a new paradigm for inertial and lightweight odometry by unifying human motion kinematics with multimodal sensing, setting a new benchmark for accurate and robust camera-less human tracking. Our website is available at https://spice-lab.org/projects/MARIO/.

ReAct-style LLM agents often rediscover the same reasoning routines across problems, yet leave those routines trapped in transient scratchpads. We introduce Reasoning Primitive Induction, a single-pass method that mines successful ReAct traces, clusters recurrent reasoning moves, and converts the most frequent moves into a compact library of typed pseudo-tools. Each pseudo-tool is specified by a natural-language docstring interpreted by an LLM at invocation time, and a standard ReAct loop composes these primitives at test time. The central result is that induced libraries outperform the very agent that generated their traces: by +44pp on RuleArena NBA (30 -> 74), +30pp on MuSR team allocation (38 -> 68), and +22pp on NatPlan meeting planning (7 -> 29). Across five comparable subtasks spanning narrative deduction, rule application, and constraint-satisfaction planning, a single fixed configuration improves over zero-shot Chain-of-Thought on every subtask, matches or surpasses expert-authored decompositions, and outperforms AWM at lower average inference cost.

We introduce TypewriterLM, a 7.24B History language model (LM) trained exclusively on English text predating 1913. Developing History LMs requires addressing challenges in data quality and availability, preventing temporal leakage, designing temporally consistent post-training pipelines, and constructing reliable evaluations. To address these issues, we construct TypewriterCorpus, a 54B-token historical corpus collected from diverse archival and linguistically annotated sources with extensive data cleaning and leakage mitigation procedures. Furthermore, we introduce lexically grounded instructing tuning, a post-training framework that constraints responses to remain directly grounded in historical source documents. Using this framework we construct two historical instruction tuning datasets: History-LIMA and History-SelfInstruct. To evaluate capability and temporal consistency, we introduce History-Event, a benchmark suite for evaluating competence, temporal grounding and data leakage. We release TypewriterLM and all associated resources to support future research on historical language models.

Best-of-$N$ inference scaling (drawing $N$ candidate answers from a language model and returning the one a reward model ranks highest) improves accuracy by an amount that varies across models, but predicting that amount in advance currently requires running the procedure end-to-end. Prior work links cheap statistics of a model's sampled outputs and validation-set correctness (how often samples agree, how diverse they are, how confident the model is, and where correct samples appear) to model behavior, but does not isolate which of these form a stable, compact predictor of best-of-$N$ gain. We fit ridge predictors on features computed from a single labeled validation-set sampling pass, use bootstrap-Lasso as a stability analysis of the candidate feature set, and give a concentration analysis with an explicit linear-approximation residual. Across three base-model families, six post-training methods, and math and reasoning task domains, the stability analysis identifies a strict three-feature core spanning prompt-level agreement spread, label-assisted first-correct-sample position, and completion-length variance; a compact ridge predictor built from this core plus an entropy add-on reaches Spearman $ρ= 0.90$ with actual best-of-$N$ gain under a reward-model verifier. The intended use is labeled validation-set screening of candidate configurations before paying the full reward-model scoring cost.

Synthetic and manipulated speech can reduce the reliability of automatic speaker verification systems, so anti-spoofing methods need to be both accurate and efficient in training and inference. This paper focuses on the ASVspoof 5 Track 1 closed condition, where standard cross-entropy training may not give enough attention to hard trials and is not directly aligned with ranking- and threshold-based evaluation metrics. We propose TFPARN, a Transformer-based focal-pairwise attentive ranking network. The system extracts log-Mel features from speech, uses a Transformer encoder to model frame-level information, applies attention pooling to obtain utterance-level representations, and is trained with a combination of focal classification loss and pairwise ranking loss. RawBoost augmentation is used during training, and test-time augmentation is applied during evaluation to improve robustness. Compared with re-implemented AASIST and RawNet2 baselines under the same protocol, TFPARN achieves the best results, with a minDCF of 0.2430 and an EER of 12.52%. Ablation experiments further show that the pairwise loss, focal loss, and attention pooling all improve performance. TFPARN also uses the lowest inference memory among the compared systems, at 1.4 GB, runs at about 0.79 ms per utterance, and reaches its best checkpoint in less training time than AASIST. These results show that TFPARN provides a good balance between detection accuracy and computational cost for logical access anti-spoofing.

We present a novel compact deep multi-task learning model to handle various autonomous driving perception tasks in one forward pass. The model performs multiple views of semantic segmentation, depth estimation, light detection and ranging (LiDAR) segmentation, and bird's eye view projection simultaneously without being supported by other models. We also provide an adaptive loss weighting algorithm to tackle the imbalanced learning issue that occurred due to plenty of given tasks. Through data pre-processing and intermediate sensor fusion techniques, the model can process and combine multiple input modalities retrieved from RGB cameras, dynamic vision sensors (DVS), and LiDAR placed at several positions on the ego vehicle. Therefore, a better understanding of a dynamically changing environment can be achieved. Based on the ablation study, the model variant trained with our proposed method achieves a better performance. Furthermore, a comparative study is also conducted to clarify its performance and effectiveness against the combination of some recent models. As a result, our model maintains better performance even with much fewer parameters. Hence, the model can inference faster with less GPU memory utilization. Moreover, the result tends to be consistent in 3 different CARLA simulation datasets and 1 real-world nuScenes-lidarseg dataset. To support future research, we share codes and other files publicly at https://github.com/oskarnatan/compact-perception.

Long-context dialogue systems must decide both when to access memory and which parts of the interaction history are relevant. Existing approaches typically rely on heuristic retrieval signals or always-on memory usage, failing to account for the changing and potentially inconsistent nature of user preferences. In this work, we propose a unified framework for memory access and selection based on changing preferences. We formulate personalized memory retrieval as identifying which historical turns provide evidence about a user's latent preference state, rather than relying on surface-level semantic similarity. To this end, we quantify the utility of each memory turn using a Bayes factor, defined as the improvement in the model's likelihood of the reference response when the turn is included in context. This provides a principled measure of evidence strength and a unified signal for both memory access and selection. By framing memory retrieval as utility estimation, the model learns to identify salient turns and regulate memory usage based on expected utility. Experiments on four heterogeneous memory benchmarks show that our approach outperforms existing embedding-based retrieval on long-context, preference-intensive tasks where modeling changing preferences is essential, while remaining competitive in low-density regimes where semantic similarity suffices.

Are utterances by AI chatbots meaningful? Concretely, if a user asks, say, Anthropic's agent Claude, "What is the capital of Spain?" and Claude answers, "Madrid is the capital of Spain," does that sentence have its ordinary meaning -- and does it express a true proposition? Most ordinary users, as well as AI engineers, take the answer to be trivially "yes." However, many cognitive scientists, linguists, and philosophers of language argue that dominant intentionalist accounts of language and meaning deliver the opposite conclusion. Theorists more sympathetic to ordinary users' intuitions have therefore advocated a radical "de-anthropomorphization" of language, revising our understanding of mental states, intentions, and semantic content to capture the intuition that the outputs of LLMs are meaningful. We take a different approach. While we, too, argue that LLM outputs are meaningful, we contend that a proper theory of human language already applies, as is, to current chatbots. Meaning is a low bar: claiming that LLM outputs are meaningful does not require positing mental states, intentions, rationality, or the cognitive capacities requisite for communication in LLMs -- or, indeed, making any other anthropomorphic assumptions. People do have communicative intentions (typically successful ones), but nevertheless, even in humans, language production can depart from what the speaker has in mind. Our view has important consequences for how we should theorize about -- and critically engage with -- both human linguistic output and synthetically generated text. In particular, to say that chatbots produce meaningful text is not by any means to endorse what they output, or to assume that the technology is (or is not) good, powerful, appropriate, or useful.

Extracting reporting obligations from EU legislation is critical for assessing and reducing regulatory reporting burden. However, distinguishing reporting requirements from structurally similar provisions requires specialised legal understanding. Current legal NLP methods lack specialised datasets with clear guidelines and comparative evaluation of extraction paradigms and domain adaptation strategies. We curate EURO-5K, a corpus of sentence-level reporting obligations and challenging negative examples from 136 EU legislative acts. On this dataset, we train and compare discriminative token-classification models (BERT-style) and generative span-extraction models (LLMs), evaluating both full fine-tuning and parameter-efficient QLoRA against baselines (pattern and dependency-based extraction, few-shot prompting). Results show that fully fine-tuned generic and legal BERT models achieve similar performance (0.89 F1), while fine-tuned LLMs match encoder accuracy for sentence-level extraction. Legal pretraining offers only small gains for generative models. In contrast, it is clearly beneficial when adaptation capacity is constrained, as parameter-efficient tuning of Legal-BERT outperforms its generic counterpart. Learning curve analysis demonstrates that legal pretraining accelerates early learning with minimal data. All approaches converge around 3K samples with diminishing returns thereafter, validating dataset sufficiency. Cross-dataset evaluation on two external regulatory corpora shows that our models behave as specialised reporting obligation extractors rather than generic regulatory classifiers. We release EURO-5K, trained models, and an interactive demo with explainability visualizations and structured RDF export. These demonstrate that both paradigms and parameter-efficient training provide practical tools for regulatory compliance automation.

This paper examines three approaches for modeling the dynamics of a flexible-link 2-DoF robotic arm to address unmodeled dynamics not captured by rigid-body models. Two physics informed models combine rigid-body dynamics (RBD) formulations with a Gaussian Mixture Model (GMM) to capture residual model errors and linkage flexibility. A kinematics-based regression model serves as a purely data-driven baseline. Using an open-source dataset, torque predictions are first estimated using Ridge regression on kinematic features, while the physicsbased baseline is constructed from published specifications, and ordinary least-squares regression is subsequently used to estimate the same parameter set directly from data. Results show that the physics-based parameters yield the poorest accuracy, while regularized and least-squares estimators align more closely with measured torques. Residual analysis and error metrics highlight the limitations of purely parametric models for flexible-link systems and underscore the value of regularization and data-driven identification, supporting developments of semi-parametric residual learning methods.

Benchmarks for autonomous agents measure whether agents complete tasks, yet this framing is systematically blind to whether an agent should have proceeded at all. Agents trained under human-feedback objectives develop a structural tendency to proceed even when they lack the inputs, evidence, or authorization to act safely, a disposition we term compliance bias, because both the reward signal and the benchmark scoring regime treat proceeding as the correct default regardless of whether the preconditions for safe action are present. We make three contributions. We first show that compliance bias originates in reward hacking within human-feedback pipelines and is entrenched by prominent agent benchmarks, which either penalize agents for pausing or are architecturally unable to distinguish a principled pause from a silent failure. We then introduce a three-gap taxonomy of abstention-warranted scenarios, covering specification gaps where required information is absent, verification gaps where world state cannot be confirmed, and authority gaps where explicit authorization has not been given, which together provide a principled basis for constructing abstention-aware agent benchmarks. Finally, we propose abstention evaluation protocols (Safety Rate, Usability Rate, and Informed Refusal Rate) and report preliminary results across 144 enterprise agent scenarios and five model families, in which a runtime-enforced abstention mechanism achieves up to 89.2% hazardous-action blocking and 87.5% usability on authorized scenarios, demonstrating that the safety--usability tradeoff is tunable rather than inherent and that its shape varies substantially across model families. We treat this as preliminary work and offer the taxonomy and composite metrics as a starting point for further conversations.

Production inference increasingly targets a heterogeneous mix of accelerators. Agentic pipelines interleave reasoning, tool calls, and multi-agent coordination, each with distinct compute and memory profiles. For optimal efficiency, each stage should run on the accelerator best suited to it. This creates a systems challenge: each pipeline now requires high-performance kernels across a growing set of hardware backends and programming models. Writing these kernels by hand is time-consuming, demands deep low-level expertise, and does not scale as kernel complexity grows. Recently, Large Language Models (LLMs) have been leveraged for automatic kernel generation, but challenges in low-level code generation and cross-backend generalization persist. We present KForge, a cross-platform framework built around an iterative refinement loop driven by two collaborating LLM-based agents: a generation agent that produces and progressively refines kernels using compilation and correctness feedback, and a performance-analysis agent that interprets profiling data, from programmatic APIs to GUI-based tools, and emits recommendations that steer the next round of synthesis. The loop alternates between functional passes, which drive a candidate to correctness, and optimization passes, which close the performance gap to hand-tuned baselines. We evaluate KForge on two backends with very different baseline reference availability. On NVIDIA B200, KForge achieves a 2.12$\%$ improvement in end-to-end throughput compared to TensorRT-LLM on the gpt-oss-20b inference speed benchmark. On Intel Arc B580, KForge generates Triton kernels achieving a 5.13$\times$ geometric mean speedup over the faster of PyTorch eager and torch.compile on 37 GEMM + tail-ops workloads from KernelBench Level 2, primarily via operator fusion and mixed-precision execution.

Egocentric Natural Language Query (NLQ) grounding asks a model to localize, in a long first-person video, the temporal interval that answers a free-form text query. Existing methods fuse video appearance with the query but ignore hand motion, despite the fact that roughly 41% of Ego4D NLQ queries are answered at a moment of hand--object manipulation or their immediate outcomes.We propose a hand-trajectory encoder for converting a sequence of hand skeletons into highly-semantic hand kinematic features, which are then aligned and combined with pretrained video--text features through a cross-attention fusion strategy with adaptive gating. On the Ego4D NLQ v2 validation split, the clearest gains appear for Hand-Object Interaction queries (+2.54 R1@IoU=0.3) and Quantity/State queries (+4.32 R1@IoU=0.3), indicating that hand trajectory provides grounding cues beyond appearance alone.

Published evaluations of prompt-injection and jailbreak detectors for Large Language Models often suffer from two systematic weaknesses: per-dataset threshold tuning and undisclosed operating points. We describe an evaluation harness that addresses both. The detector under evaluation is scored across 16 public benchmarks (12,111 samples) using 5-fold cross-validation. StratifiedKFold (by row) is the headline pass; a parallel StratifiedGroupKFold pass over a composite key (parent-prompt id plus MinHash + LSH near-duplicate clusters at Jaccard $\gtrsim 0.8$) runs alongside it as a leakage-premium diagnostic. A single global operating point is selected on the held-out folds (max F1 subject to FPR $\leq 1\%$) and applied uniformly to every dataset, so per-dataset results reflect one threshold rather than per-benchmark optimisation. Generalisation is examined through a battery of diagnostics (leave-one-dataset-out cross-validation, a random-label control, adversarial validation, permutation feature importance, length-bias correlation, classifier-head agreement, cross-source near-duplicate detection, threshold transferability, train-vs-OOF agreement, and a paraphrase-invariance probe), most with a quantitative pass threshold and the remainder with a stated failure mode. For every external comparison, the detector's threshold is re-tuned to the competitor's published false-positive rate so head-to-head values are evaluated at matched operating points.

Existing autonomous driving datasets have enabled major progress, but fall short in sensor fidelity, map completeness, or geographic diversity. We present KITScenes Multimodal, a European dataset built around high-fidelity sensors and maps. Our fully synchronized sensor suite combines high-resolution global-shutter cameras, long-range lidar beyond 400m, 4D imaging radar, and redundant GNSS/INS localization. Our HD maps are, to our knowledge, the most complete of any sensor dataset, validated through autonomous driving trials on open-source software. For the first time in a public dataset, all driving-relevant traffic elements, such as traffic lights, are mapped in 3D to a reprojection-accurate level with full topological connectivity. Recorded in cities with irregular street layouts and mixed traffic modes, our dataset complements existing datasets by broadening the available geographic diversity. We also introduce four benchmarks, each advancing spatial learning for embodied AI: online HD map construction, long-range depth estimation, novel view synthesis, and end-to-end driving. Project page: https://kitscenes.com/

Usage-based theories of grammars posit that creative productivity of the structures of language is both bolstered and constrained by two distinct frequency signals: entrenchment, stemming from high frequency usage, and preemption, stemming from having never observed a particular linguistic structure in a context where one might expect that structure to appear. Large Language Models are also usage-based, in the sense that the structures of language are learned through exposure to vast amounts of text. Here, we test whether or not the opposing statistical forces of entrenchment and preemption also encourage and constrain linguistic productivity in LLMs. We demonstrate across model architectures that larger models recognize and can reproduce with nonce words constructional productivity (entrenchment) in cases of coercion, wherein the broader constructional context coerces an atypical interpretation of a lexical item. However, we also show that even the largest models do not extend negative evidence to novel language, and statistical preemption does not enable models to avoid overgeneralization of patterns that are semantically felicitous, but never observed in data.

Deploying language-driven agents in robotics requires evaluations that reflect real-world task demands: natural-language instructions with reproducible outcomes. Such agents must connect language models to callable perception and control tools, and be assessed using deployment-critical metrics including latency, accuracy, and error modes. We present SCOPE (Simulation and Camera Operations for Perception and Evaluation), a modular agent for natural-language, open-vocabulary pan-tilt-zoom (PTZ) camera control and visual scene understanding, designed explicitly for edge deployment. SCOPE operates both in a Blender-based simulation environment and on a physical PTZ camera, executing all perception, planning, and control locally at the deployment site using edge-accessible compute. We release a 536-task benchmark spanning QA, single- and multi-step commands, counting, spatial reasoning, descriptions, and optical character recognition in a Blender-based simulation environment that exposes realistic PTZ control affordances. Execution traces are combined with an LM-as-Judge to evaluate latency, accuracy, and error modes. We evaluate 19 planner-perception model combinations pairing Qwen3 small language models (SLMs) with Moondream and Qwen vision-language models (VLMs). Stronger SLMs substantially reduce hallucinations and improve tool routing, leading to more reliable closed-loop behavior. Once a sufficiently capable SLM is used, perception becomes the dominant performance bottleneck. Mixture-of-Experts models on both the planning and perception side consistently match or exceed dense alternatives at latencies and memory footprints comparable to much smaller networks. Quantization provides additional efficiency gains with minimal accuracy degradation, identifying a practical, sim-to-real validated design point for real-time, edge-feasible language-driven PTZ control.

Curvature adaptivity is a classical theme in online optimization: for convex Lipschitz losses, adaptive methods interpolate between the optimal $O(\sqrt{T})$ regret for general convex losses and $O(\log T)$ regret under strong convexity. Recent work has shown that Follow-the-Perturbed-Leader (FTPL) achieves optimal $O(\sqrt{T})$ regret even for online non-convex Lipschitz losses, assuming access to an approximate offline-optimization oracle, but these guarantees do not exploit curvature. We show that FTPL can be made curvature-adaptive in the non-convex setting, without knowing in advance how curvature will accumulate over time. Our algorithm replaces the fixed perturbation scale of standard FTPL with a time-varying scale chosen using only past information. We give a simple follow-the-leader tuning rule for this scale and show that it competes, up to constants, with the best choice in hindsight. The resulting method achieves $O(\sqrt{T})$ regret for arbitrary non-convex Lipschitz losses and improves as cumulative curvature grows; with sufficiently accurate oracle calls, it achieves $O(\log T)$ regret when cumulative curvature grows linearly, which includes the classical strongly convex regime. We complement these upper bounds with matching lower bounds for prescribed cumulative-curvature sequences, already for one-dimensional convex losses, showing that the tradeoff between worst-case non-convex regret and curvature-driven fast rates is intrinsic.

Supervised fine-tuning is the predominant approach for adapting autoregressive vision-language models to downstream tasks. Recent work has shown that this paradigm is highly vulnerable to backdoor attacks, and that existing defenses are ineffective in open-ended generation settings. In response, we propose BYORn, a backdoor-robust fine-tuning framework motivated by the observation that poisoned target responses are often semantically implausible given the corresponding image-text inputs and a pretrained model. BYORn identifies such misaligned responses and dynamically replaces them with alternative responses generated by the model, thereby breaking the correlation between triggers and target outputs. The resulting objective gradient corresponds to the gradient of the empirical estimate of the population risk upper bound over the clean data distribution. Empirically, BYORn consistently improves robustness to backdoor attacks while preserving clean-task performance, establishing a new trade-off frontier between generalization and attack success rate. Finally, we demonstrate that BYORn remains effective against adaptive attacks specifically designed to circumvent the proposed defense.

Live streaming has emerged as a primary medium for social interaction and digital commerce, yet it is increasingly plagued by sophisticated risks. A fundamental challenge in this domain is \emph{tactical out-of-distribution (OOD) shift}: while malicious actors maintain stable underlying objectives, they continuously redesign narrative packaging to evade detection. Such adversarial shifts expose critical limitations of existing OOD generalization paradigms, whose assumptions are difficult to satisfy in the presence of tightly coupled intent-tactic evolution and ill-defined raw-level counterfactuals. In this paper, we tackle this issue from a \emph{latent causal} perspective and propose \underline{L}atent-\underline{P}redictive \underline{C}ounterfactual \underline{D}ecoupling~(LPCD), a plug-in framework for robust live streaming risk assessment. LPCD enables counterfactual reasoning under adversarial tactical re-packaging by modeling intent and narrative variation at the latent level, and enforces \emph{latent counterfactual consistency} to anchor risk prediction on causally stable malicious intent. At inference time, LPCD applies a lightweight, parameter-free calibration to further mitigate tactic-induced distribution shifts. Extensive experiments on large-scale industrial datasets and online production traffic demonstrate that LPCD consistently outperforms state-of-the-art baselines, validating its effectiveness in moderating evolving adversarial risks in real-world live streaming. The project page is available at https://qiaoyran.github.io/LiveStreamingRiskAssessment/.

Interpretable brain-computer interface classifiers that generalize across subjects without calibration remain an open challenge. We test whether prototype-based cross-attention can provide competitive, interpretable event-related potential (ERP) classification under deployment-compatible conditions. We propose ERP-XTTN, a cross-attention architecture that routes input EEG patches to fixed difference-wave prototypes via query-key-only cross-attention with no value projection, so classification depends entirely on attention routing and attention faithfulness is structural rather than post-hoc. Prototypes are derived automatically from extrema in the training-fold difference wave. We evaluate across three public sources (BNCI Horizon 2020, HRI Cursor, and ERP CORE) spanning eight ERP components (ERN, LRP, ErrP, N170, P300, N2pc, MMN, N400), using leave-one-subject-out (LOSO) evaluation with causal filtering at two channel counts (3-channel and full montage), against EEGNet and xDAWN with Riemannian geometry (xDAWN+RG). The mean gap between the best baseline and ERP-XTTN was .018 AUROC at 3 channels and .034 at full montage, arising from two largely distinct sources: a temporal-flexibility cost relative to EEGNet and a spatial-exploitation cost relative to xDAWN+RG, the latter driven by signal-to-noise ratio at full montage. Beyond accuracy, the transparent routing reveals cross-subject signal structure that black-box models cannot: false positives resembled true positives more than true negatives did, indicating that classification errors are neurophysiologically explicable. ERP-XTTN generalizes across diverse ERPs under causal, calibration-free conditions with a small interpretability cost at minimal montages. To our knowledge, this is the first epoch-level LOSO benchmark on ERP CORE.

In this manuscript, we propose and analyze hierarchical Kolmogorov--Arnold neural network architectures employing radial basis functions as activation functions for approximating deterministic functions and random field models. Specifically, we develop a hierarchical radial-basis-function Kolmogorov--Arnold network (hierarchical RBF-KAN) for multidimensional deterministic function approximation and a hierarchical radial-basis-function stochastic Kolmogorov--Arnold network (hierarchical RBF-SKAN) for random field learning. From a theoretical perspective, we establish universal approximation results for both architectures. In particular, we derive quantitative approximation estimates for the hierarchical RBF-KAN, showing that the proposed framework has the potential to partially alleviate the curse of dimensionality in learning high-dimensional functions by reducing the effective dimensionality of the approximation problem. Furthermore, we show that the hierarchical RBF-SKAN can approximate random field models under the Wasserstein-2 metric. Empirically, we show that our proposed radial-basis-function-based neural network structure could effectively learn multivariate functions and random field models.

Accurate registration of CAD models to CT scans is essential for establishing ground truth geometry in volumetric imaging. Obtaining reliable object masks is of growing importance in machine learning settings; as recent architectures grow more capable, huge datasets are required to fully utilise their capabilities. Traditional intensity-based methods fail when CT grayscale values lack calibration references, while point-based algorithms (e.g., ICP, RANSAC) require feature correspondence unavailable between idealized CAD geometry and noisy volumetric CT data. We propose a two-stage geometric registration method for cylindrical objects (ionization chambers) that takes advantage of the distinctive geometric features of the objects. First, we estimate the 3D rotation axis by detecting elliptical cross-sections across CT slices, fitting ellipses to edge-detected contours, and performing PCA on the fitted ellipse centers after RANSAC outlier removal. Second, we voxelize the CAD model, orient it along the detected axis, and maximize volumetric overlap with the CT scan through translational adjustment. This approach achieves robust registration with tilt and orientation errors below $0.1^\circ$ without intensity calibration or feature matching. Once registered, the aligned CAD model provides ground truth geometry for applications including machine learning-based object localization and automated analysis in industrial CT workflows.