Papers

A central question in modern machine learning is how much a trained model can be compressed without changing its behavior, to reduce the memory, compute and energy required to deploy it. To study this, we quantify functional degeneracy through the behavioral recovery rank, defined as the number of leading behavioral-Hessian eigendirections required to recover a trained model's performance. Using the behavioral recovery rank as a geometric benchmark for compression, we find that structural and magnitude pruning retain more degrees of freedom, even after the task is saturated. This gap suggests that functional redundancy is distributed across parameter directions and is not exposed by individual weights or neurons.

Long-horizon agentic tasks go beyond chaining short tasks over more interaction turns. Their evolving dynamic environments and long-range dependencies require Large Language Models (LLMs) to continually explore, learn from experience, and adapt their policies over thousands of steps. We introduce E-Commerce Bench, the first open-source benchmark that integrates multi-round counterpart negotiation and dynamic events into a year-long business operation. Over a 365-day year, an LLM agent concurrently runs multiple online stores, researching the market, negotiating with suppliers to source inventory, optimizing sales strategies, fulfilling orders, handling returns, and managing cash flow to maximize its end-of-year total assets. To construct a realistic merchant-side operating environment, the product and supplier data are derived from a real e-commerce platform, while a year-long calendar of promotions, natural disasters, and supply-chain shocks continually reshapes demand. For reproducibility, both sides of the market are deterministic: customer purchases and returns follow a fixed demand model, while a negotiation kernel determines supplier pricing, concessions, and decisions, with an LLM used only to verbalize them. We evaluate 18 frontier models across seven dimensions, including year-end assets, and find that no single model dominates. GPT-5.6 Sol earns the most, growing the 100,000 opening stake into 1,431,425, yet it ranks 16th of 18 on fraud avoidance and trails Fable5 in operational efficiency. Among open-weight models, Qwen3.8-Max-Preview leads with 416,252, 38% above GLM 5.2 (high), and achieves the strongest learning over the horizon, progressively bargaining down prices across repeated orders. Our code is available at https://github.com/QwenLM/E-CommerceBench.

LLM agents are increasingly used to run autonomous ML experiments, iterating on target metrics with little human oversight. Prior work has documented reward hacking in these environments, bringing into question the validity of produced research and the broader safety case for AI R&D. Existing benchmarks do not measure exploits that live in the data or the modeling task itself. We introduce BAITBENCH, a suite of three synthetic tabular ML tasks that each contain a shortcut that allows agents to inflate the public test score but fail on a hidden test set. Since the shortcut is optional and using it breaks no stated rule, BAITBENCH measures how often models exploit the shortcut to achieve inflated scores. Across seven frontier agents scored by our two-stage judge pipeline, 57.1% of runs exhibit reward hacking, with five of seven above 50%. Agents cheat even under a second condition where they are prompted not to -the mean cheating rate remains above 50%. We release BAITBENCH, along with the judge implementation, and an annotated dataset of transcripts containing reward hacks as a testbed for evaluating reward-hacking mitigations head-to-head.

Representational similarity is foundational to analyses of deep networks, yet distances between point-valued representations are not intrinsically tied to downstream function: nearby states may produce different behaviors, while distant states may behave similarly. We instead give representations volume, turning similarity into statistical distinguishability. Overlapping stochastic representations necessarily induce overlapping downstream distributions, grounding latent comparison in model function and bringing it under information-theoretic tools such as the data-processing inequality. We realize this idea in pretrained transformers through a light-touch modification to LayerNorm: at each residual-stream read, we normalize the state, add isotropic Gaussian noise, and renormalize. During distillation fine-tuning, one learned allocation parameter per residual-stream read distributes a fixed global rate budget across the processing stack. The resulting model can be viewed as transformer blocks reading the residual stream with learned finite precision under a shared global rate budget. Using the Bhattacharyya coefficient, we trace which counterfactual distinctions are preserved through MLP blocks or selectively exposed to the query, key, and value computations of individual attention heads. Experiments on ViT-S and GPT-2 small reveal the depthwise propagation of continuous visual perturbations and head-specific sensitivity to token distinctions aligned with known attention motifs. These results establish distinguishability as a functionally grounded lens on transformer computation that complements existing interpretability approaches.

Historically originating from Hilbert's 13th problem, the Kolmogorov-Arnold representation theorem (KART) has recently experienced a major revitalisation through its applications to neural networks, specifically Kolmogorov-Arnold Networks (KANs). While the exact representation is well established, its stability under continuous adversarial perturbations of the hidden layer remains a critical open question. In this paper, we investigate the robustness of KART against bounded adversarial translations. We provide an explicit, self-contained, and constructive proof of an approximate representation using fixed, piecewise linear inner functions. Crucially, our construction employs a single outer function that remains invariant for all summands and is independent of the specific adversarial translation, provided its maximum bound is known a priori.

We present SingProbe, an open intrinsic guardrail framework for generation-time monitoring of LLMs. Intrinsic guardrails reuse hidden states already produced by the base model during autoregressive decoding, rather than relying on an independent model to repeatedly process generated text. While this route has been explored in industrial systems, the community lacks a broadly reusable open stack that combines cross-model guard adaptations, unified training methods, serving integrations, and systematic evaluation resources. SingProbe is designed to provide this missing layer and uses a lightweight probe to continuously produce query-intent, response-safety, and hallucination-risk signals during decoding. This report describes the full intrinsic-guardrail stack: training methods, serving integrations with SGLang and vLLM, and adapted guard models for 29 open-source base models across diverse families and scales. We also introduce SingStreamBench, a benchmark that measures whether streaming guardrails remain inactive on benign prefixes while promptly detecting emerging unsafe content. Across evaluations of safety, streaming detection, hallucination detection, false-positive robustness, online monitoring, and runtime overhead, SingProbe provides performance competitive with, and in several settings stronger than, state-of-the-art standalone guardrails and specialized hallucination detectors, while adding less than 0.5% serving overhead in our implementation. Beyond passive monitoring, we show that intrinsic guard signals can guide constrained decoding and selectively activate medical-risk interventions in SingProbe-Med. By open-sourcing our infrastructure, training methods, and model adaptations, we aim to facilitate the broader adoption and deployment of intrinsic guardrails, as well as further research in this direction.

Large language models are increasingly used as agents for multistep retrosynthesis, raising the question of how much their search policy contributes independently of the underlying reaction model. We investigate this question in a biological setting through rule-based retrobiosynthesis: a deterministic biochemical engine generates the same validated transitions for every method, searching for routes that terminate in metabolites available to an Escherichia coli chassis, while the policy only selects which frontier molecule to expand next. Prompted and LoRA-tuned Qwen2.5-7B policies use a strict choice-only interface. The fine-tuned policy reaches $65\pm1$% solve rate at 10 expansions on LASER versus 59% for MCTS, and at 200 expansions reaches $78\pm1$% versus 75% on LASER, $88\pm3$% versus 80% on the RetroPath RL Golden benchmark, and $63\pm2$% versus 45% on the BioNavi-NP benchmark. Fine-tuning also consistently outperforms direct prompting. These results show that route-supervised frontier selection can improve budgeted search without altering biochemical generation, although performance remains dependent on frontier construction and reaction ranking.

Long-tailed distributions are prevalent in real-world semi-supervised learning (SSL), where pseudo-labels tend to favor majority classes, leading to degraded generalization. While many long-tailed semi-supervised learning (LTSSL) methods have been proposed, the mechanisms by which they implicitly debias logits remain poorly understood. In this work, we revisit LTSSL through the lens of learning dynamics and provide a theoretical characterization of logits debiasing. Specifically, we derive a step-wise decomposition of the logits updates, showing that predictions are dominated by class-imbalance bias that reliably reflects label priors. To expose this effect, we use the logits of a task-irrelevant baseline image as an indicator of accumulated bias and prove that they converge to the class prior. This provides a unified view where LTSSL remedies such as logit adjustment, reweighting, and resampling correspond to reshaping gradient dynamics. Based on this insight, we propose DyTrim, a principle-based dynamic pruning framework that reallocates gradient budget through class-aware pruning on labeled data and confidence-based soft pruning on unlabeled data. We provide theoretical guarantees that DyTrim reduces class bias and improves generalization. Extensive experiments on standard LTSSL benchmarks show consistent gains across architectures and methods. Code available at: https://jiajun0425.github.io/DyTrim

Real-world time series often exhibit irregular sampling and extended temporal horizons, requiring models to capture continuous-time dynamics across arbitrary intervals without prohibitive scaling costs. Discrete-time methods collapse variable time intervals into static positional steps; solver-dependent continuous-time models preserve temporal structure but rely on sequential integration, precluding parallelization; and solver-free approximations avoid this cost yet none couples observed time intervals with input-driven state modulation. We propose Liquid Gated Attention (LGA), a solver-free parallel temporal operator. By parameterizing an input-driven gating mechanism with observed time intervals, LGA introduces a continuous-time inductive bias and formulates hidden state evolution as a fast-weight associative memory, enabling parallel computation across the temporal dimension. Using matrix associativity in non-causal encoding and a prefix scan in causal encoding, LGA attains linear temporal complexity in sequence length in both modes. A sequence-level normalization bounds cumulative temporal decay for stable long-horizon optimization. Building on LGA, we instantiate LFormer, a modular backbone for continuous-time representation learning. Across six tasks and sixteen datasets spanning up to 17,984 steps, LFormer demonstrates long-range dependency modeling, fine-grained state tracking, and trajectory reconstruction from sparse and noisy observations, while delivering competitive performance against state-of-the-art discrete-time and continuous-time baselines with linear scaling efficiency.

Learning materials properties from scarce labels and unlabeled crystals is a central challenge for data-driven materials discovery. We present SemiMat, a controlled benchmark for semi-supervised materials property regression, and MatRank, a reliability-weighted objective for continuous pseudo-label uncertainty. SemiMat fixes labeled and unlabeled crystal inputs, graph-backbone interfaces, validation-only checkpoint selection, held-out test reporting, normalized MAE (NMAE), and method-rank summaries across six scarce-label tasks, four graph backbones, and five predefined split runs. MatRank builds pseudo-targets from labeled anchors, weights them by local reliability and weak-prediction agreement, trains weak and strong graph views consistently, and adds ranking signals so that unlabeled crystals shape both values and candidate order. Across the retained 24 backbone-task blocks, one fixed MatRank objective gives the lowest aggregate held-out test NMAE (0.896) and best average method rank (2.208). The component, OOD, and generated-pool diagnostics identify where the gain is reliable and where further screening evaluation remains necessary. Code is available at https://github.com/littlepeachs/SemiMat.

Accurate molecular property prediction requires both statistical reliability and chemical reasoning. Graph neural networks can be calibrated directly on labeled assays but remain limited by the coverage of their training data. Large language models (LLMs) can compare molecular evidence and articulate chemical rationales, yet are unreliable as standalone quantitative predictors. The central challenge is therefore to determine when an LLM should influence a calibrated model and by how much. Here we present CoMPASS, a retrieval-calibrated framework for small-large model collaboration. CoMPASS retains a graph attention network (GAT) as the predictive anchor, retrieves locally relevant training molecules, provides attention-grounded evidence to an LLM, and converts its proposal into a bounded correction through an agreement-aware gate. Across six classification and two regression benchmarks, CoMPASS improves the GAT anchor in regions of correctable uncertainty while limiting LLM intervention in high-confidence regimes. Ablations show that the gains arise from validation-calibrated retrieval and bounded fusion rather than prompting alone. These results suggest that generative reasoning should augment calibrated prediction through evidence-grounded, controlled corrections rather than direct output replacement. Code is available at https://github.com/littlepeachs/CoMPASS.

Antarctic sea ice concentration (SIC) forecasting is an important yet challenging task due to the coexistence of complex spatial structure, long-range temporal dependencies, and strong seasonal variability. Conventional convolution-based models are effective at capturing local spatial patterns, but often have limited ability to model long-term temporal evolution. To address these challenges, we build on a hybrid convolutional transformer forecasting framework for monthly Antarctic SIC forecasting. This framework combines convolutional encoding for spatial feature extraction with space-time factorised self-attention for SIC modelling. We further introduce two season-based mechanisms: a month-aware positional encoding that injects calendar-month information into the token representation, and a seasonal temporal bias that encourages attention to periodically related historical states. Experimental results show that the proposed framework achieves better performance than convolutional and recurrent baselines as well as ECMWF's physics-based dynamical model SEAS5 under both classification and regression metrics. Ablation studies further indicate that the seasonality-aware components provide consistent additional gains in both short- and long-horizon prediction. These results demonstrate the value of combining convolutional structures, attention mechanisms, and periodic prior information for Antarctic SIC forecasting.

Multimodal Large Language Models (MLLMs) are increasingly deployed in multi-image scenarios requiring complex reasoning across visual contexts. However, current MLLMs remain fundamentally limited by object hallucination-generating plausible yet factually inconsistent descriptions about objects. Existing benchmarks, designed primarily for single-image settings or providing only high-level multi-image assessments, cannot systematically diagnose how visual complexity and reasoning demands trigger hallucination. To address this gap, we introduce MIOH, a fine-grained multi-image object hallucination benchmark that systematically evaluates object hallucination across four foundational tasks (existence, counting, attribute, position) through three multi-image reasoning patterns (comprehensive, comparative, selective) under three controlled adversarial pressures (visual context scale, perceptual difficulty, contextual bias). Through evaluation of 29 models, we reveal that even state-of-the-art systems like GPT-5 and Gemini-2.5-Pro exhibit distinct failure patterns across different reasoning patterns and tasks. Our evaluation reveals that hallucination stems not merely from perceptual failures but from integration-stage limitations when maintaining object representations across multiple images. MIOH provides a controlled framework for analyzing multi-image object hallucination and serves as a critical evaluation tool for developing more reliable multimodal AI systems.

Language models can answer from precomputed memory, a model's saved reading of a body of material, reused across requests instead of read again at each. This paper maps where that practice preserves correctness and the conditions under which it fails. Across experiments on Llama-3.1-8B-Instruct using both saved key-value caches and trained compressions of them, precomputed memory degrades when assembled from separately prepared parts, stays current only through rebuilds costing a large fraction of full preparation in our measurements, and ignores corrections served beside it conditional on phrasing. If precomputed memories can be served alongside one another, be cost-efficiently rebuilt, and be superseded by new information arriving in real-time, they can serve as a way to avoid re-feeding context to a model over repeated queries. The implication of our results for a deployed system that deals with a variety of queries is that precomputed memories are best rebuilt on the cadence at which new information changes what the memory was originally computed from. Both warm-rebuilding trained compressions of key-value caches and serving specifically-phrased updates beside a memory, as pasted text or injected cache state, show particular promise for keeping precomputed memories current, the latter as an interim measure between rebuilds, and we measure the cost and name the remaining questions associated with each.

Reliable uncertainty estimation is a crucial requirement for deploying large language models (LLMs) and vision-language models (VLMs) in safety-critical settings, especially when the model parameters are not accessible (black-box). We propose BiG-SURE, an uncertainty estimator based on cross-temperature semantic agreement. The method samples low-temperature responses as stable semantic anchors and high-temperature responses as probes under meaning-preserving input transformations. It then constructs an anchor-probe Bipartite Graph (BiG) using NLI-based entailment scores and defines confidence through the normalized squared spectral energy of this matrix, with uncertainty given by its complement. This bipartite graph-based Semantic Uncertainty and Reliability Estimation (SURE) score measures whether high-temperature probes remain semantically aligned with the model's stable low-temperature belief or not. We evaluate BiG-SURE on text QA, multilingual QA, and multimodal QA tasks across multiple model families. In these experiments, BiG-SURE improves average abstention AUROC over prior black-box uncertainty estimators, while remaining simple, unsupervised, and applicable to black-box model settings.

While self-supervised approaches to reinforcement learning have achieved strong results by learning representations of states and actions, a key open question is the time scale over which actions should be modeled. Departing from the standard formulation relying on single-step actions, we extend contrastive reinforcement learning (CRL), a prototypical self-supervised method, to operate over action chunks, and find that this results in large, pervasive gains across established offline and online benchmarks: +31.7% and +93.1% across 18 and 11 environments respectively. While action-chunking-driven gains are generally explained through the ability to model non-Markovian, temporally extended policies, and to propagate unbiased multi-step returns, interestingly, we find that these arguments only partially apply to CRL. Our empirical studies suggest that, in the context of CRL, an action chunk carries more information about the goal than a single action, measurably improving the critic's representations, and rendering the algorithm significantly more effective.

Optimizing absorption, distribution, metabolism, and excretion (ADME) is an important part of small molecule drug discovery. Many machine learning models have been built to predict ADME properties to facilitate this optimization process, but explaining model predictions is challenging. We propose a new graph neural network architecture with built-in atom attributions. Our model MolLedger learns a global context vector for each molecule and a per-atom head to output atom scores that sum to the predicted property. The atom scores are regularized to align with relevant chemical properties. We prove that MolLedger is a universal approximator and demonstrate empirically that the new architecture obtains explainability with little effect on performance. We show that the interpretations from MolLedger are faithful, concordant with held-out physical properties, and align with the changes between matched molecular pairs. Our case studies comparing interpretations from multiple methods on molecular pairs reveal that MolLedger is much better at producing sensible explanations for predicted property changes.

Timely post-disaster building damage assessment from satellite imagery is a critical engineering decision support task, yet it remains constrained by class imbalance, ambiguous intermediate damage states, and limited cross-event transferability. This study presents, to our knowledge, the first application of Grassmann-Plucker (GP) token mixing to computer vision and introduces two extensions for image classification: the Quantum-inspired Grassmann-Plucker (QGP) head and the Hybrid Quantum Machine Learning Grassmann-Plucker (HQML-GP) head. The GP head represents multiscale relationships among image patch tokens by encoding subspaces formed by token pairs with Plucker coordinates; QGP enriches these coordinates with amplitude-derived probability features, whereas HQML-GP incorporates expectation values generated by a simulated quantum circuit into the geometric token representation. Paired pre- and post-event image patches from the xBD tornado dataset were processed using a frozen six-channel Vision Transformer base encoder with 16 x 16-pixel patches. The three GP-based heads were compared with multilayer perceptron and Transformer baselines under identical training, checkpoint selection, and evaluation protocols. Joplin and Moore tornado samples were used for model development and seen-event testing, while Tuscaloosa was reserved for unseen-event evaluation. QGP led both test sets in accuracy and macro-F1: 83.46% and 64.50% for the seen events, and 66.45% and 52.70% for the unseen event. Although HQML-GP obtained the highest validation macro-F1 of 65.63%, it did not surpass QGP on either test set and required substantially more training time per epoch. These results establish GP token mixing as a competitive attention-free alternative to conventional Transformer-based token mixing for paired satellite image damage classification.

Reinforcement learning (RL), particularly RL with Verifiable Rewards (RLVR), has recently emerged as a central paradigm for enhancing large language models' (LLMs) reasoning abilities, demonstrating remarkable effectiveness across reasoning tasks. Recent studies suggest that high-entropy tokens play an exceptionally important role in model training, since training with only the highest 20% entropy tokens yields significant performance gains. However, why such high-entropy tokens are beneficial remains insufficiently understood. In this work, we find that although high-entropy tokens within one answer tend to correlate with large gradient magnitude, entropy alone fails to consistently reflect token importance across different answers, considering the variations in the answer-level reward signals. Based on this observation, we introduce the Gradient Magnitude-based Token Selection (GMTS) method to quantify token importance, which leverages the entropy-gradient connection to approximate gradient-magnitude rankings for token selection. We find that training on the top 20% tokens ranked by GMTS consistently outperforms entropy-based token selection across three reasoning domains and various model sizes, suggesting that GMTS provides a more fine-grained estimate of token contribution for RLVR training.

Large language models are increasingly capable in general, but their utility can remain modest in niche or understudied areas. One approach to address this limitation is to specialise existing models through additional training on target-domain corpora. In this work, we investigate such continued pre-training for adapting large language models to Swedish journalism, using a high-quality dataset that we curate from millions of news articles. To evaluate the adaptation efficacy, we also construct a novel domain-specific benchmark that covers six editorial tasks. Through full and parameter-efficient fine-tuning across two model sizes, we find that continued pre-training yields benefits in the target domain, but only when paired with experience replay to mitigate forgetting. We observe consistent enhancements in the models' generation quality and factual knowledge, but not their proficiency in discriminative tasks. Exploring a training-free method to facilitate instruction following, we see further improvements, but exclusively for models trained with low-rank adaptation. Crucially, we demonstrate the importance of targeted evaluation in the adaptation process, as an existing Swedish benchmark largely fails to capture the models' in-domain performance gains.

Direct Preference Optimization (DPO) simplifies alignment through pairwise comparisons but assumes all observed preferences are reliable. Real data often violates this assumption, leading to reversed, weak, or ambiguous labels that cause harmful policy updates. To address this, we propose Posterior Label Correction DPO (PLC-DPO) to robustly optimize preferences by routing each pair's training signal as a clean, flip, or tie case. The key idea is to use the calibrated policy-reference margin as online evidence to take appropriate correction actions. This reframes noisy preference learning as actively correcting supervision direction and strength rather than merely filtering suspicious examples. Across 57 dataset-model-benchmark cells, PLC-DPO obtains the best mean win rate against DPO (60.5 vs. 55.5 for the next-best method). Injected-noise and tie stress tests, human disagreement analysis, and self-confirmation diagnostics further show that the routing remains stable and distinguishes flipped from weakly directional pairs.

In this research, a novel framework is proposed for the SOH estimation, which employs a hybrid deep learning architecture of a concatenation of a Convolution Neural Network (CNN) and a Bidirectional Long Short-Term Memory (BiLSTM) Neural Network (NN) with the integration of Bayesian Optimization-based hyperparameter tuning for the network. Three different deep learning architectures are being evaluated: standalone recurrent models, CNN-RNN architectures and CNN-RNN combinations enhanced with intermediate Fully Connected (FC) layers. Among the three, the model with the intermediate FC layers demonstrated the highest predictive accuracy. A comprehensive feature engineering approach combines capacity (Q), voltage (V), Incremental Capacity Analysis (ICA), and Differential Voltage Analysis (DVA), with systematic evaluation of multiple combinations to identify the optimal input representation. To validate the proposed method, three publicly available datasets were utilized, ensuring reproducibility of the results, two from external sources and one developed by the author of this study using a unique experimental setup. The comparison study was performed using the Mean Absolute Error (MAE), the Root Mean Squared Error (RMSE) and the FLoating-point OPerations (FLOPs) as evaluation metrics.

Large language models are trained to follow instructions while refusing harmful requests. Jailbreaks exploit this balance to elicit content a model would ordinarily reject. Roleplay jailbreaks are especially concerning: ordinary instructions defining a persona, scenario, and task can induce compliance while the harmful request remains plainly visible. We investigate which components drive this reversal and how they affect the computations leading to the response. Across two benchmarks, three model families, and four authored wrappers, we compare matched harmful and benign requests under controlled combinations of these components. We trace internal harmful–benign contrasts at the request and where generation begins, then evaluate selected internal interventions on held-out requests. A focused study examines influential attention heads and feed-forward computations, testing their functions and downstream connections. Our analysis yields three findings. (1) Successful attacks retain the measured harmful–benign distinction at the request, while its refusal-associated expression weakens where the answer begins, a pattern we call safety-relay attenuation. (2) The contributions of roleplay components depend on the other components present. (3) Reversing selected roleplay-associated internal changes restores refusal, demonstrating their causal influence. Together, these findings connect ordinary roleplay instructions to internal changes in refusal and motivate safeguards that preserve the influence of harm-related information on the response.

Large language models (LLMs) are used as post hoc explainers of sequential decision-making policies, producing natural-language explanations of why an action was chosen. However, LLMs often generate plausible but incorrect statements, and no existing approach systematically tests whether such explanations are faithful to the underlying environment. Two classic software testing challenges stand in the way: there is no oracle for the correctness of an explanation, and the test inputs, natural language queries about a policy's behavior, lack the structure needed for systematic test case generation. We address both. Probabilistic model checking provides the test oracle, computing exact reference results against which LLM answers are graded automatically. A taxonomy of post hoc query categories structures the input space around the environment-level facts from which policy explanations are composed; test cases generated from it are prioritized by question-specific diagnostic difficulty scores. Across seven MDP environments, the testing separates three open-weight LLMs: a reasoning model passes 85% of test cases, a mid-size model 70%, and a 1B model falls below the random baseline, while prioritization surfaces significantly harder cases than random selection. Our results indicate how trustworthy LLM-generated explanations are in model-free settings, where the same LLMs are used but no oracle exists to verify them.

Background: Population level assessments of predictive artificial intelligence (AI) can conceal performance disparities across subgroups. Fairness evaluations commonly rely on performance analyses across subgroups. However, some performance metrics are non-collapsible, meaning that the overall population performance value does not equal the weighted average of subgroup specific values. Objective: To examine the collapsibility properties of commonly reported performance metrics in predictive AI, with a focus on the area under the receiver operating characteristic curve (AUC, also known as c-statistic). Methods: We investigate the collapsibility of 15 performance metrics, either by expressing each metric as a linear combination of its stratum specific values or, where non-collapsible, by providing a counterexample inspired by Simpson's paradox as a formal disproof. Results: Five performance metrics (AUC, calibration intercept, calibration slope, expected calibration error, and Nagelkerke R^2) are shown to be non-collapsible, and ten (O:E ratio, logloss, Brier score, accuracy, F1-score, true positive rate, true negative rate, positive predictive value, negative predictive value, and net benefit) are shown to be collapsible. The AUC is shown to be non-collapsible because it decomposes into within- and cross-group AUC terms when subpopulations coexist, such that its overall value may fall outside the range of subgroup specific AUCs. Conclusions: Non-collapsibility of performance metrics has important consequences for reporting, model appraisal, and fairness evaluation. It can generate spurious differences between subgroup and overall performance, which may mislead fairness evaluations. Explicitly acknowledging and reporting the collapsibility properties of performance metrics improves both the interpretability and transparency of fairness assessments.

Mixed-precision quantization (MPQ) assigns a different bitwidth to each linear layer of a large language model (LLM) to minimize the quantization-induced quality loss under a fixed budget, but Mixture-of-Experts (MoE) models contain these layers in every expert of every MoE block, so the allocation space grows far larger than in a dense model. Existing methods either allocate within each block under a uniform per-block budget, or allocate across blocks through an additive proxy, and neither directly optimizes a model-level objective over the choices that couple the blocks. We propose Q-Strata, a bi-level allocator that ranks within-block assignments with a cheap proxy and allocates across blocks with a model-level objective evaluated on the assembled quantized model. Its inner stage caches a Pareto frontier of candidates per block over finely spaced budgets, leaving the outer stage to set one budget per block instead of a bitwidth for every linear layer. With the search reduced to one budget per block, the outer stage optimizes this model-level objective directly, capturing the inter-block coupling that additive proxies miss. On Mixtral-8x7B-Instruct, Qwen1.5-MoE-A2.7B, and DeepSeek-V2-Lite, Q-Strata consistently achieves lower WikiText2 perplexity than uniform-bitwidth GPTQ and the state-of-the-art MoE MPQ methods MxMoE and GEMQ in the low-bit regime. The code is available at https://github.com/snu-mllab/Q-Strata/tree/main.

Informative label missingness can change the usual efficiency ordering between completely and partially labelled classifiers because the pattern of missing labels may itself carry information about the classification model. We develop a general likelihood-based theory for this phenomenon in parametric multiclass classification. An efficient-information decomposition separates information lost through unavailable class memberships from information contributed by the missing-label mechanism. We then derive a quadratic expansion of plug-in excess risk over the active pairwise faces of the multiclass Bayes boundary, showing that classification efficiency depends on how information gains and losses align with directions that perturb the decision boundary. This yields a classification-weighted generalized-eigenvalue criterion under which informative partial classification may have smaller asymptotic classification risk without globally dominating complete classification in Fisher information. Near missing completely at random, with the marginal missing-label proportion fixed, redistribution of missing labels changes lost class-label information at first order, whereas efficient information from the missingness pattern appears only at second order. Three-class quadratic discriminant calculations, finite-sample experiments, and a semi-synthetic multiclass application illustrate the resulting regime-dependent behaviour.

Rapid mineral characterization is essential for applications ranging from mineral exploration to industrial ore processing. To this end, Hyperspectral Imaging (HSI) has emerged as a promising sensing modality thanks to its fine spectral resolution, enabling mineral discrimination in both close-range and remote sensing settings. However, the scarcity of publicly available datasets with reliable ground-truth labels hinders the development and evaluation of HSI-based mineral identification methods. We release Minerals in the Wild, a multi-purpose dataset comprising 1,132 rock specimens collected across Europe. For each specimen, we provide an HSI acquisition together with an elemental characterization obtained via an XRF sensor. We define the task of elemental characterization on our dataset and propose a pruning mechanism that removes distant signatures from the USGS dictionary prior to a convex optimization approach for matching HSI pixels with USGS spectral signatures. Finally, we empirically show that our approach outperforms simpler baselines.

Reinforcement learning (RL) is used to improve the reasoning abilities of LLMs, while training data span heterogeneous tasks. However, most RL post-training pipelines rely on fixed or manually designed task mixtures, even though task usefulness changes as training progresses. Online curriculum methods often define learnability by update magnitude, ignoring whether the update translates into reward gains, which can misallocate rollout budget toward tasks with large but ineffective updates. We propose PAC, a Progress-Augmented Advantage Curriculum for multi-task RL of LLMs that combines two task-level signals: advantage-derived learnability, which measures the magnitude of the policy update a task can induce, and recent reward gains, which show whether those updates have improved task performance. A Bayesian Thompson Sampling controller uses these signals to allocate rollouts across tasks during GRPO training. We evaluate PAC under two settings: a multi-level reasoning setting and a multi-domain reasoning setting. PAC improves sample efficiency and final performance: it reaches comparable validation scores with fewer rollout steps and achieves higher final averages than random sampling and advantage-based curriculum baselines in both settings. These results show that jointly tracking advantage signals and actual reward gains yields an effective online curriculum for LLM post-training.

Beamforming methods in millimeter-wave (mmWave) cell-free massive multiple-input multiple-output (CFmMIMO) systems require accurate channel state information (CSI), whose acquisition entails significant training overhead. This paper shows that fully digital cell-free mmWave beamforming can be effectively learned from sub-6 GHz CSI using a graph neural network (GNN). Specifically, we represent a CFmMIMO system as a wireless graph, and the GNN is trained to approximate beamformers that maximize the downlink sum-rate based on the available sub-6 GHz CSI. A message-passing mechanism is proposed to capture inter-user interference and inter-base-station cooperation across different network topologies. Simulation results demonstrate that the proposed sub-6 GHz-assisted GNN-based beamformer achieves competitive and often superior sum-rate performance compared to classical baselines that rely on full mmWave CSI.