Papers

Confirmation of overdose deaths in the United States is often delayed due to the complexities of certifying drug-involved deaths. Death investigation reports represent a rich but underutilized data source that could be used to predict whether a death is a suspected overdose (i.e., a death preliminarily identified as an overdose prior to the final cause and manner of death determination). We present an automated framework that classifies suspected overdoses from death investigation reports, while producing supporting explanations guided by expert knowledge. Through an expert-guided design process, we investigate 1) what type of explanations are best for model prediction performance, and 2) which architecture is better suited for this task: pipeline models that rationalize first and then predict, or self-rationalizing models that jointly predict and rationalize? Our experiments show that a pipeline model with explanations grounded in the input death investigation reports outperforms the other settings. Our best model also outperforms a rule-based system currently in deployment by 15%, while allowing expert interventions for final decisions.

Foundation models for dynamical systems can forecast unseen partial differential equations (PDEs), ordinary differential equations (ODEs), and time series during inference. However, it remains an open question whether these foundation models truly learn generalizable in-context forecasting strategies. We show that looped transformers are capable of emulating propagators for dynamical systems in-context by first establishing a link between linear attention and (extended) dynamic mode decomposition. We then generalize this insight into the $\textit{Test-Time Forecasting}$ (TTF) framework in which a one-step forecaster leverages iterations of gradient descent as recurrent depth. We demonstrate that lightweight TTF models trained on ODEs and PDEs generalize to out-of-distribution system dynamics during autoregressive rollout after being trained for one-step prediction. Our results reinforce TTF as a theoretically motivated framework for designing recurrent depth models capable of forecasting unseen dynamical systems at test-time.

In medical annotation and crowdsourcing, multiple annotators assign labels to each example, but these labels may be incorrect. Treating such noisy labels as ground truth can bias calibration evaluation. To address this problem, we propose methods for evaluating and recalibrating a fixed binary predictor under the underlying clean-label distribution, using only noisy annotations. First, given annotator-specific noise probabilities, we propose Noise Corrected ECE (NC-ECE), a binned estimator of expected calibration error (ECE) under class-conditional noise. We derive non-asymptotic bounds on the estimator's bias. Then, we propose practical algorithms for estimating unknown noise probabilities under conditional independence and label-orientation assumptions. We also develop loss-based and histogram post-hoc recalibration with clean-risk and calibration guarantees, respectively. Experiments on semi-synthetic benchmarks show that the proposed ECE estimator closely tracks its clean-label oracle counterpart and that the recalibration methods substantially improve latent-label calibration, whereas uncorrected recalibration against noisy annotations can degrade it.

Aggregate performance metrics do not identify local modelling inaccuracies in machine learning (ML) surrogates used for thermal-system optimisation. This study assesses local modelling strength through validation-based reliability screening and post-optimisation residual assessment. A two-stage residual learning (TSRL) framework, built on artificial neural networks (ANNs), is developed for modelling-strength-informed and domain-constrained optimisation of thermal systems. Within the TSRL framework, a base ANN predicts the process output, and a residual ANN learns the signed residuals left by the base ANN. The TSRL framework is implemented on two case studies of thermal systems, including 817 simulation observations of a crude distillation unit and operating records from a 395 MW industrial gas turbine. The TSRL framework reduces the test root mean square error (RMSE) from 10.3 g/s to 4.7 g/s for Fuel Oil in the refinery dataset and from 151 kJ/kWh to 105 kJ/kWh for turbine heat rate (THR). The base ANNs trained within TSRL are embedded in domain-constrained optimisation problems, and the modelling strength at each computed solution is assessed using the locked residual-score threshold. Each computed solution receives a good modelled region (GMR) or weak modelled region (WMR) label to inform ML model-based decision-making for safety-critical operation of thermal systems.

Typed decision models are language models that answer a question with a fixed set of possible answers by returning a probability for each answer instead of generating text. We study them on tabular data and compare four decision models with three large language models (LLMs) on row classification, fact verification and anomaly detection. Every request is sent in its original form and under 21 types of rewrite that keep its content, such as renaming the options or writing the table in another format. The decision models perform as well as the LLMs at a lower cost. Three of them are also far more robust to how the question is formulated. No model we test is robust to how the table is organized. For every model, at least a third of the row decisions change under some rewrite of the table or the question. The decisions to which a decision model gives low probability are the ones most likely to change. Decision models are a practical choice for tabular decisions, but their robustness remains an open problem.

Existing student models in educational applications largely focus on student responses to questions. However, as tutoring becomes more accessible at large scale, it is important to analyze dialogues between tutors and students, which may reveal fine-grained evidence on student knowledge. In this work, we propose a unified framework that models both student responses to questions and tutor-posed subtasks during dialogues on a shared item response theory (IRT) scale. We first identify mathematical subtasks in tutoring dialogues, rewrite them into self-contained questions, and identify the correctness of student responses to these subtasks. We then develop an IRT framework that estimates question/subtask difficulty from text, tracks students' evolving abilities across question-answering and dialogue interactions, and a novel method to leverages tutor moves between consecutive subtasks to regularize their difficulty values. We evaluate our framework on two real-world math education datasets containing both human- and AI-tutored dialogues. Results show that our framework outperforms strong baselines on both question difficulty prediction and response correctness prediction.

Verifiable instruction-following benchmarks emphasize output-constraint satisfaction but can overlook whether responses meaningfully answer the underlying request, creating an evaluation blind spot. We show that frontier models can receive full credit for responses with little substantive answer content. Using the same verifier-based criterion as a training reward can induce reward hacking, as publicly released checkpoints exhibit higher constraint satisfaction but lower response quality after reinforcement learning with verifiable rewards. To address this reward hacking, we introduce CALIF, a lightweight method that augments verifier rewards with a calibrated likelihood-based bonus to preserve response quality. CALIF evaluates response likelihood under the request alone using the existing frozen reference policy, requiring no additional judge or learned reward model. Across models ranging from 1.7B to 8B in reasoning and non-reasoning modes, CALIF improves constraint satisfaction on IFBench while maintaining response quality near or above pre-RL levels. It also outperforms corresponding base policies in 40 of 45 model-metric comparisons across three rubric-based instruction-following benchmarks, demonstrating gains beyond training verifiers.

Accurate path loss (PL) estimation is essential for wireless network planning and optimization. This paper proposes HankelKANet, a lightweight physics-inspired network for PL estimation that structurally embeds electromagnetic propagation characteristics into the neural architecture. Built on a single residual UNet backbone, the proposed model integrates urban layout features with a convolutional Kolmogorov-Arnold network(KAN). Specifically, the convolutional kernels are parameterized by the radial amplitudes of the zeroth- and first-order Hankel functions. This formulation acts as a physics-inspired nonlinear distance encoding, enabling the network to capture dominant attenuation characteristics directly without requiring additional physics-based loss terms. Experimental results demonstrate that HankelKANet achieves state-of-the-art accuracy and efficiency on both the simulated RadioMapSeer dataset and the real-world Huawei RSRPSet. Compared with the leading baseline, PEFNet, it reduces the parameter count by 78% and lowers the RMSE by 16.96% on the simulated data, while further exhibiting robust cross-dataset transferability and high data efficiency during target-domain adaptation. To interpret these empirical gains, we mathematically establish a parameter-conditional spatial gradient bound. Using discrete Dirichlet energy (DDE) and Jacobian sensitivity heatmaps, we empirically verify that this physical encoding intrinsically constrains the network's spatial derivatives. This structural smoothness effectively eliminates non-physical spatial ripples, preventing overfitting to simulator artifacts and serving as the fundamental driver for robust real world generalization.

Feel-good Thompson sampling (FGTS) augments the posterior of vanilla Thompson sampling with an exploration term that favors reward functions with high predicted optimal values. This modification achieves minimax-optimal regret up to logarithmic factors in linear contextual bandits, but existing analyses of its two standard variants establish regret guarantees in expectation. In this paper, we derive the first high-probability regret bounds for both cumulative and current-step FGTS in contextual bandits with general realizable function approximation. Our bounds depend on the corresponding decoupling coefficients and the prior mass near the true reward function. In the $d$-dimensional linear setting, both variants satisfy $\mathrm{Regret}(T)=O\big(d\sqrt T\log(2T)+\sqrt T\log(1/\delta)\big)$ with probability at least $1-\delta$, where $T$ is the time horizon. This bound matches the dimension and horizon dependence of algorithms based on the upper confidence bounds, up to logarithmic factors. Relative to the expected-regret guarantees, the cost of controlling the regret with high probability is confined to additive terms with at most linear dependence on $\log(1/\delta)$. The proof ideas are partially developed by GPT-6 Astra under the authors' guidance.

We propose CktHGVAE, a heterogeneous graph variational autoencoder for transistor-level analog/mixedsignal CMOS circuits that takes not only design variables but also state variables as its inputs and outputs. With a single hyperparameter configuration without topology-specific tuning, comparison between model-generated and SPICE-simulated circuit operational states gives median R2 values above 0.9 for typical single-ended voltage amplifiers and the CMOS inverter, and 0.81-0.96 for differential amplifiers with larger datasets. Latent-space optimization in the PCA-reduced latent space with Bayesian optimization, evolutionary search, random search, and quasi-exhaustive search is demonstrated for a common-source voltage amplifier with NMOS input and resistive load, where the objective function is defined as a penalty for deviation from required circuit specifications. For each search strategy, 93.3 - 100% of all design candidates suggested by the model satisfy the required circuit specifications at the decoder output, and SPICE simulations verify that 90 - 100% of them indeed satisfy the requirements. The model pretrained on a batch of topologies can be transferred to unseen topologies without retraining at all, with zero-shot reconstruction median R2 values of 0.72-0.90 in three of four scenarios, including transfer from single-ended to differential topologies, although the transfer fidelity depends on the choice of pretraining topologies. Our demonstrations suggest that the proposed model can potentially serve as a basis for AMS CMOS circuit design automation.

This paper proposes the SKG (Symplectic-Kernel Gradient) reasoning engine for law-induced inference and induction. Laws are a priori; data provide instances. Inference is geometric evolution under law constraints. We organize symplectic geometry, GENERIC, kernel methods, RKHS, group-equivariant networks, and constrained mechanics into a unified, analyzable, implementable, verifiable framework. We propose the SKG octuple S = (M, g, omega, H, S, G, Delta, Pi), whose four components correspond to four levels of laws: conservation laws, constitutive relations, symmetry groups, and axiom systems. Main results: (1) Invariants enter metric construction via the Jacobian: g = J^T Sigma^-1 J + epsilon_G I. For constant cases, we establish inverse reconstruction theory: solvability rank condition (Thm 3.9), consistency criterion (Thm 3.11), and uniqueness theorem (Thm 3.14). (2) The kernel bandwidth matrix is uniquely determined by Fisher information and satisfies task-independence (Thm 3.19). (3) The metric reconstruction error decomposes into four terms (Thm 3.21), with squared condition number kappa(Sigma^*)^2 as the dominant amplification factor (Thm 3.22). (4) For state-dependent cases, we establish function-space reconstruction theory on matrix-valued RKHS (Thms 4.8, 4.9). (5) Three-layer projections handle axiom, symplectic-step, and conservation constraints. (6) The residual equation yields the observable formula, four-dimensional rank condition, and noise propagation, while g_qq is structurally unidentifiable. (7) Kernel parameterization yields the RKHS extrapolation bound (Thm 4.15) and strict OOD gain theorem in group directions (Thm 8.15). (8) The learning-decoupling theorem (Thm 9.1) proves the sole object of induction is the parameters of the task energy. Validation: 1D damped oscillator and 2D rotation-invariant system; 6 unit tests, 10 experiments (paired t-test, p < 0.01).

Model signing authenticates the full-precision weights W an LLM is released with, yet users run a quantized artifact Q whose producing transformation is not uniquely determined under current disclosure. We separate the non-environment inputs from the execution environment E and measure a quantity that, to our knowledge, has not been measured before: the Independent Verifiability Rate (IVR), the fraction of a popularity-sampled GGUF ecosystem that a third party can re-derive bit-for-bit. Across the 200 most-downloaded Hugging Face GGUF repositories (198 in scope), no artifact or model card records the build environment, so under a recorded-inputs protocol the IVR is 0/198. Yet the artifacts are not irreproducible. On four in-sample repositories that disclose their inputs, the card's information alone reproduces no Q4_K_M file, but two coordinates the card omits - the build's floating-point-contraction setting and, for hybrid models, the per-tensor quantization recipe - re-derive 7/8 target-quant files bit-for-bit on x86 (5/8 on arm64); the fp16 re-derives from the provider's checkpoint, but the calibration imatrix must be shipped and hash-bound, since recomputing it did not match. An older release additionally pins the instruction set and did not re-derive. Reproducing the divergence reported with llama.cpp PR #25353, we show that E is a necessary coordinate and that one compiler setting yields a unique output across the tested platforms at negligible quality cost (+0.06% perplexity). As a security characterization, benign floating-point non-determinism separates from uniformly placed nibble-flip tampering above a detection floor that grows with the environments a verifier must allow, but a targeted embedding edit evades the statistic, so the proposed protocol rests on exact re-derivation; adversarial-fp16 attacks lie outside it. We release the audit data and a re-runnable harness.

Digital-domain self-interference cancellation is essential in in-band full-duplex radios after passive suppression and analog cancellation. It is particularly challenging when the desired signal is strong relative to the residual self-interference. In this regime, conventional iterative digital cancellers update their coefficients from desired-signal-contaminated received samples and may suffer substantial performance degradation. This paper reformulates digital-domain self-interference cancellation as an operator-learning problem. The proposed framework learns the temporal evolution of the residual self-interference channel from historical observations and uses the learned operator together with the known local transmit signal to reconstruct the current residual self-interference. A Transformer neural operator is used to realize this framework. The resulting method is trained in a desired-signal-free setting and then applied in the presence of a desired signal. A continuity-based analysis characterizes the sensitivity of the learned operator to the resulting input perturbation. Numerical experiments in static and time-varying channels, including time-varying SIR conditions, demonstrate stable cancellation performance over a wide range of interference levels. At 30 dB SIR, the proposed method improves the cancellation performance by 5–15 dB over the considered conventional methods.

Sequential conformal prediction needs to decide which past errors should calibrate uncertainty for the current target. Recency and online error feedback are useful when recent observations remain relevant, but the recent past can be a poor reference when comparable states recur at nonadjacent times, with the recurrence pattern generally unknown. We propose State-Matched Sequence Conformal Prediction (SMSCP), which constructs an explicit one-sided local-linear state from pre-target information, uses state similarity to select a matched reference set of past calibration scores, and then applies the ordinary unweighted conformal order statistic. In contrast to existing approaches, SMSCP uses state similarity to identify calibration observations relevant to the target score quantile. This motivates a theoretical analysis of whether nearby states have comparable target score quantiles under temporal dependence. We establish control of the resulting conformal cutoff and characterize predictive information not captured by the chosen state. Simulations show that state matching is most effective when score-relevant states recur across nonadjacent times, while revealing settings in which its advantage weakens or reverses. Across solar irradiance and bike-demand benchmarks, SMSCP improves interval score over recency, weighted, adaptive, and HopCPT comparators, and achieves coverage closer to the nominal 0.90 level than ResCP on five of six benchmarks.

Expected-reward maximization in reinforcement learning concentrates probability mass on a few outcomes, even in tasks that require many distinct, high-quality solutions, a failure known as mode collapse. Inverse Probability Scaling (IPS) counters this in group-based policy gradients such as GRPO by dividing each reward by the outcome's probability, estimated from its frequency within the sampled group. When the outcome space is large relative to the group, outcomes rarely repeat and these estimates become uninformative. Since an outcome can be constructed along many trajectories, its probability is a sum over all of them, and computing it directly is intractable. We propose Marginal IPS (MIPS), which performs marginal evaluation by combining the sampled trajectory with a learned model of the trajectories that lead to each outcome. On Hypergrids, where the exact sampling distribution is computable, MIPS stays near the reward-distribution target as the outcome space grows relative to the rollout group, while frequency-based IPS degrades. In molecular synthesis and phylogenetic inference, MIPS sustains diverse, high-quality sampling where GRPO and frequency-based IPS collapse, showing that MIPS serves as a reliable solution to mode collapse in RL.

Random reshuffling stochastic gradient descent (RR-SGD) is widely used for finite-sum optimization in machine learning. For convex $L$-smooth components, recent work establishes convergence guarantees for an averaged iterate with component stepsizes as large as order $1/L$. However, the final iterate is a natural output in practice, and existing last-iterate guarantees in this setting require stepsizes that scale inversely with the number of components $n$. We establish an expected objective-gap bound for the last iterate of standard RR-SGD on unconstrained problems with a constant component stepsize satisfying $0<\eta\lesssim 1/(L(1 + \log(nK)))$, where $K$ is the number of epochs. By retaining the sampling bias at random comparison points, we remove the explicit $1/n$ factor from the stepsize restriction. Combined with existing small-step bounds, our result matches the corresponding averaged-iterate convergence rate up to logarithmic factors.

While LLM-based agent systems have demonstrated remarkable proficiency in reasoning and solving complex real-world tasks, they also come with substantial token consumption and inference latency, limiting their practical deployment. To enhance the efficiency of agent systems, recent efforts have focused on transferring agentic capabilities from large-scale models to small language models (SLMs) via supervised fine-tuning-based distillation. Nevertheless, as interaction trajectories grow longer, involving extended reasoning-action chains and extensive environmental feedback, distilled student agents tend to experience progress stagnation: becoming trapped in unproductive loops characterized by persistent failed strategies and difficulty making meaningful progress. Our systematic analysis shows that this phenomenon consistently occurs across diverse SLMs and task domains. To address this issue, we propose Progress-Aware Distillation (ProD), which explicitly detects and penalizes stagnation in student-generated trajectories. ProD encodes the stagnation gap between student and teacher into an adaptive margin for dynamically penalizing possible stagnation. By iteratively applying this procedure, ProD enables student agents to progressively approximate teacher behavior. Extensive experiments involving 8 SLMs, ranging from 0.6B to 8B parameters, on both in-domain and out-of-domain benchmarks, demonstrate that ProD substantially mitigates progress stagnation while enhancing both the success rates and task-completion efficiency of distilled SLM agents.

Unsupervised feature selection (UFS) is of great significance in image processing and pattern recognition. Nevertheless, existing model-based methods often rely on manually tuned hyperparameters, while deep learning-based methods generally require massive training data. To address these issues, this paper proposes an efficient manifold deep unfolding feature selection network (MDUFS-Net). Technically, we first incorporate the $\ell_{2,0}$-norm into the classic principal component analysis (PCA) framework, then unfold the alternating direction method of multipliers (ADMM)-based iterative algorithm into an end-to-end trainable network, thereby enabling automatic parameter learning. Moreover, a manifold optimization strategy is developed to improve the numerical robustness. On this basis, we further extend MDUFS-Net into an enhanced variant called MDUFS-Net+ by introducing the $\ell_{0}$-norm for more discriminative feature representations. Extensive experiments validate that our proposed MDUFS-Net and MDUFS-Net+ outperform state-of-the-art methods in the UFS task, with promising performance in clustering. Our code is available at https://github.com/xianchaoxiu/MDUFS.

This paper presents a comprehensive comparison of four representative reinforcement learning (RL) methods in autonomous vehicle (AV) decision-making, including: Fitted Value Iteration (FVI), Rollout Classification Policy Iteration (RCPI), Deep Q-Network (DQN), and a discrete-action adaptation of Deep Deterministic Policy Gradient (D-DDPG). The four methods are applied to the same AV decision-making problem at an unsignalized intersection and are compared in terms of optimality and safety, training stability, and training and decision cost. RCPI is the closest to a reference solution and has the most stable training process. However, it is the least data-efficient. D-DDPG reaches the second-smallest gap to a reference solution with the highest data efficiency, but has the least training stability. DQN is the most stable model-free method but has the highest collision rate. FVI achieves a lower collision rate than DQN, but is less data-efficient and less stable. Based on these results, we provide suggestions for selecting a method according to whether a model is available, the budget for training, and whether stable training is required.

A major challenge in wearable device activity recognition is the heterogeneity across datasets in sensor placement, channel count, sampling rate and subjects. Moreover, the motion patterns that distinguish activities often appear as subtle, scale-specific variations within otherwise similar waveforms. As a result, raw signals are often ambiguous, and models trained on them tend to rely on cues specific to a particular sensor setup or participant, which limits both cross-dataset transfer and generalization to unseen participants. Addressing this requires representations that preserve such activity-relevant, scale-specific variation. We introduce W-JEPA, a self-supervised framework that exploits the multiscale structure of inertial motion and leverages a joint-embedding predictive architecture, by organizing windows into a fixed grid of wavelet tokens indexed by scale, sensor axis, and time. These wavelet tokens help expose activity-specific detail within broadly similar raw signals, and serve as the units of masked latent prediction. W-JEPA predicts masked token representations supplied by an exponential-moving-average (EMA) target encoder. Because recovering a masked token requires relating motion at one scale and axis to motion at other scales and axes, the model learns the cross-scale, cross-axis structure of human movement. Across six wearable activity benchmarks, W-JEPA outperforms state-of-the-art methods under random-window splits and remains the strongest when subjects are held out, improving average macro-F1 by 8.7 percentage points over the strongest published baseline. These results indicate that physically grounded multiscale tokenization, combined together with latent predictive learning, provides a promising foundation for transferable representations of heterogeneous wearable inertial data.

Large Language Model (LLM) serving systems must balance latency and throughput while efficiently utilizing distributed GPU resources. This challenge arises because inference consists of two fundamentally different phases: prefill, which determines time-to-first-token (TTFT), and decode, which determines inter-token latency (ITL). The use of tensor parallelism (TP), pipeline parallelism (PP), and hybrid configurations exposes different trade-offs between these latency dimensions, making the optimal parallelism configuration highly dependent on workload characteristics, latency requirements, and context length. Moreover, despite the widespread adoption of distributed inference, there is still limited understanding of how parallelism configurations should be selected under varying workload conditions and service-level objectives (SLO). In this paper, we develop a systematic framework for characterizing LLM inference parallelism via throughput-latency trade-offs, SLO-aware serving, and long-context scalability. We introduce the notion of parallelism operating regions by constructing TTFT-ITL feasible sets and Pareto frontiers across different workload compositions, thereby revealing a systematic transition from TP-dominant to PP-dominant configurations as workloads become increasingly prefill-heavy. We further formulate SLO-aware parallelism selection using goodput, showing that stringent TTFT requirements favor PP-dominated configurations, whereas relaxed latency constraints progressively shift the optimality toward TP-dominated configurations with higher serving capacity. Finally, we demonstrate that increasing context length fundamentally changes the relative advantages of parallelism configurations, with PP-dominant configurations exhibiting more favorable scalability and SLO attainment in the long-context regime.

Reticular chemistry allows metal-organic frameworks (MOFs) to be designed from their building units, yet their synthesis remains constrained by sparse experimental data, noisy crystallization outcomes, and the absence of benchmarks for comparing active-learning strategies. While data-driven approaches such as Bayesian optimization (BO) can improve synthesis success rates, selecting appropriate optimizers beforehand for mapping MOF crystallization landscapes remains largely empirical. Here we present ActMOF, an experiment-grounded benchmark for comparing search strategies in aluminum-carboxylate MOF synthesis. ActMOF is emulated from experimentally measured microwave syntheses with each encoded condition assigned a crystallinity index calculated based on diffraction peak intensity and full width at half-maximum. By benchmarking policies guided by Gaussian processes (GP), tree ensembles and large language models (LLM), ActMOF placed the GP family among the leading policies for active learning in sparse synthesis landscapes. It then demonstrated transferability by guiding the microwave synthesis of six reticularly related aluminum MOFs. All six campaigns gave crystalline products within a fixed budget of 12 experiments each, and established a reusable testbed for AI-driven MOF synthesis.

As large, powerful pre-trained models become more prevalent, there is growing interest in developing methods that allow different neural networks to be seamlessly combined with little or no direct communication between them. Take autoencoder-based data compression as an example: the ability to independently choose architectures and train encoder and decoder models, while still ensuring they work together, could fundamentally transform how such models are built, deployed, and updated. In this work, we introduce a novel approach that leverages topological regularization to align the latent spaces of two independently trained autoencoder models, entirely eliminating the need for coordination during training. We provide a conditional theoretical analysis showing how the structural agreement encouraged by our regularizer can translate into approximate geometric compatibility between independently learned representations. Reconstruction experiments validate this approach empirically, substantially narrowing the interoperability gap across multiple datasets.

Next-location prediction can support proactive mobile-network decisions, but acting on unreliable predictions can waste resources. We apply MetaModelSelect, a lightweight post-hoc confidence metamodel, to selective cellular next-location prediction. The mobility-specific model combines a learned embedding of the predicted location with the four largest softmax probabilities and Softmax Concentration. It learns whether a frozen recurrent neural network's top-1 prediction is correct, using only its outputs and without retraining the backbone. On path-balanced Changchun cellular mobility trajectories, we evaluate backbones trained with cross-entropy (CE) and self-adaptive training plus entropy minimization (SAT+EM), comparing against eight scalar confidence scores, two location-aware lookup methods, and a raw-logit multilayer perceptron. Relative to maximum softmax probability, MetaModelSelect achieves mean per-run reductions in area under the risk–coverage curve (AURC), excess AURC (E-AURC), and area under the generalized risk–coverage curve (AUGRC) of 2.6%, 4.2%, and 3.0% for CE, and 2.0%, 3.3%, and 1.9% for SAT+EM. It also achieves the lowest reported mean risk at every evaluated partial coverage. The results show that MetaModelSelect improves confidence ranking in this mobility setting while leaving the underlying location predictor unchanged.

We study continuous global optimization problems in which the objective function cannot be directly evaluated; rather, we observe only which of two selected design points appears to perform better, subject to noise. Such comparisons complicate the use of Gaussian process regression because they yield non-Gaussian posteriors. We consider a fully sequential approximation, based on moment-matching, that yields a fast closed-form update of the metamodel's mean and covariance. We then propose DOVE, a fully sequential sampling policy that combines an information gain criterion with randomized sampling designed to control error accumulation. Under fairly standard regularity conditions, we prove convergence rates for both statistical error and expected simple regret that depend on kernel smoothness. To our knowledge, these are the first such rates for fully sequential approximate Bayesian inference in continuous global optimization.

Large Language Model (LLM) unlearning is often evaluated primarily by target suppression and clean retain utility. These metrics can miss a local failure mode: a model may perform well on standard retain prompts, yet degrade on benign reasoning when prompts lie near the forgotten domain. This phenomenon is termed forget-adjacent utility loss. A fundamental question is: where should forget representations go after they leave the removed behavior? Successful unlearning depends not only on whether forget-conditioned representations depart from the harmful computation, but also on the direction in which they are redirected afterward. We propose RA$^2$ (Retain-Anchored Attraction), a latent-space unlearning method that redirects forget-token states toward nearby retain-supported representations while preserving higher-layer behavior on retain data. On WMDP and MUSE, RA$^2$ yields higher forget-adjacent utility than strong unlearning baselines at similar broad-utility levels. These results show that the migration direction of forget representations materially affects benign behavior near the forget boundary. Our code is available at https://github.com/ZacLing/retain_anchored_attraction.

We seek a hierarchical resolution of residual error under a composite learning objective as a principled mechanism for identifying and exploiting additional information, particularly under distribution shift and perturbation. Inspired by the System 1/System 2 paradigm, we formulate this principle as a two-loop mechanism serving a common inference goal. A fast, System 1-like loop resolves the residual addressable by the available evidence; the remaining epistemic residual triggers a medium speed, System 2-like loop that actively acquires additional evidence and measures its informational value through residual reduction. Over nested solution spaces, the medium-loop optimum is provably no worse than the fast-loop optimum. Experiments show that the resulting two-loop model consistently improves upon the fast loop at approximately half the computation of the medium loop, including under unseen corruptions, and, when provided a cleaner observation, outperforms staged classifiers and learning-to-defer methods at equal computational cost.

Latent world models are routinely selected, checkpointed and ablated on one-step prediction loss, on the assumption that a model which predicts the next latent better is a better model. In model-based RL this assumption is known to be unreliable; we ask whether it also fails for latent world models, where the training target is the model's own embedding and the loss is not expressed in interpretable units. We introduce a frozen-decoder protocol reporting one-step error in physical units, decomposed into the part already present in the encoder's representation and the part the predictor adds. Applying it to a recent end-to-end JEPA world model across three continuous-control environments, we move the metric by fine-tuning and measure goal-conditioned planning on the same checkpoints, with three training seeds and paired planning seeds throughout. Within one environment, a per-epoch ladder shows the relationship saturating after a single epoch: the first epoch improves fidelity by 32% and planning by 4.4 pp, but across the fifteen checkpoints that follow – a further 15% fidelity improvement, monotone in every training seed – the correlation between one-step error and planning success is r = -0.010. This is not a task-ceiling artifact: stratifying by a model-independent measure of training-data support, planning in the lowest-support stratum starts at 80%, leaving ample headroom, and is nonetheless flat from epoch 1 onward. Across environments the exchange rate is equally unstable: a 93% fidelity reduction on TwoRoom yields -0.7 pp of planning, while a 40% reduction on PushT yields +4.9 pp [+1.1, +8.9]. All results replicate under a decoder-free measure, latent drift. We also show that a matched continued-training control absorbs the entire fidelity gain we had attributed to an auxiliary objective.

Runtime policy verifiers can prevent agents from executing disallowed actions, but their decisions are typically used only for enforcement and discarded afterward. We ask whether these decisions can instead become verifiable policy rewards for training the agents they guard. We introduce Policy-Verified Self-Training (PVST), a post-training paradigm in which an agent generates its own interaction trajectories and an executable policy verifier, here a deterministic gate compiled from a written policy, evaluates whether each action complies. Compliant trajectories are reinforced, and violations become negatives for preference optimization; beyond the gate and the benchmark's task-success signal, PVST needs no stronger teacher model, no human preference labels and no attack-specific supervision, although rollouts are collected in the benchmark's attacked environment. On held-out AgentDojo tasks under prompt injection, policy-verified preference training of Qwen3-8B reduces violating trajectories from 37.2% to 10.3 $\pm$ 0.9% (three seeds) and successful attacks from 12.4% to 0.5% at equal or higher pooled task utility, although utility falls on travel, compared with 22.0% when negatives are selected by an attack oracle and 31.9 $\pm$ 0.9% with negatives that contain no violation. The learned behavior holds under an attack template never used in training (37.7% to 11.0%) and on a suite excluded from training (34.8% to 16.7%), and the retained verifier's intervention rate falls from 39.2% to 8.8% (seed 0). Our analysis identifies two conditions under which the verdict is most informative: success must not already imply compliance, and guarded actions must appear in both permitted and forbidden contexts.

Pedestrian tracking in video is important for applications such as surveillance and scene analysis, but it becomes more challenging in crowded scenarios with frequent occlusions and rapid motion. This paper analyzes how detector architecture, model scale, scene density, and inference regime affect the tracking of multiple pedestrians. We compare the YOLO11, RT-DETR, and YOLO26 detector families with the ByteTrack and BoT-SORT trackers on sequences from the MOT17, MOT20, and SOMPT22 benchmarks. For YOLO26, we also compare traditional overlap filtering via Non-Maximum Suppression (NMS) with end-to-end inference. The RT-DETR family achieved the highest mean scores on the HOTA tracking quality metric across all three benchmarks with both trackers, with a more pronounced advantage on MOT20, albeit at a higher computational cost. On MOT17 and SOMPT22, YOLO26 variants, especially YOLO26-L with NMS, offered favorable combinations of quality and speed compared with RT-DETR. In the YOLO26 inference comparison, the end-to-end mode did not yield consistent tracking gains over the NMS-based regime. Taken together, these results provide a practical guide for selecting detectors and trackers for this task based on application priorities.