Papers

Modern distributed training is costly to debug, tune, and analyze because faithful analysis often requires repeated target-scale executions. We observe that many analysis tasks depend on the original distributed execution path, but not on full all-rank execution. We present RankLens, a single-rank emulation system that runs one survivor rank on a single GPU while preserving the target job's distributed semantics. The survivor rank executes the original training program and performs real local computation. At communication boundaries, the system replaces real multi-rank communication with task-dependent semantics, selectively restoring recorded communication results and recorded or estimated completion timing while preserving the asynchronous dependencies required for analysis. This enables replay from real executions and supports analysis, debugging, and tuning with existing debuggers and performance analysis tools. We evaluate the system on DeepSpeed and Megatron-LM across 8-64-GPU target configurations. RankLens achieves 100% training-tensor communication-structure match, keeps emulation timing error below 5%, supports local anomaly reproduction and execution analysis, and reduces the resource requirement of these analyses from target-scale clusters to a single GPU.

LLM agents now act as delegated communication intermediaries: they read private context, draft messages, and send externally visible emails on behalf of a user. Existing checker-repair defenses can reduce inappropriate disclosure, but they rely on repeated calls to a remote action LLM (the model that drafts and repairs the email) and a remote checker LLM whenever a draft fails validation. This creates a cost problem as well as a privacy problem, especially when a requester strategically asks for exact values, raw records, source evidence, or other details that induce leakage-heavy drafts. We propose local contract-conditioned rewriting as a cost-efficient enforcement layer for email-generating agents. The mechanism intercepts an outgoing draft, constructs an information-release contract that specifies what to preserve, generalize, and forbid, rewrites the draft with a local Gemma rewriter, and selectively invokes a final PrivacyChecker. Across extensive email tasks under normal prompts and attack prompts that operationalize strategic information requests, Gemma + Checker reduces normal-prompt leakage from 11.41% to 4.03% while improving UsableRate (satisfied or partially satisfied outputs) from 85.2% to 92.6%. Under the strategic attack-prompt condition, it reduces leakage from 28.19% to 18.79% and reduces checker-only cost amplification from 1.32x to 1.02x. Offline replay ablation shows that the full contract reduces leakage relative to generic Gemma rewriting by 36.3% under normal prompts and 54.1% under attack prompts. These results support a local-first privacy enforcement design: move repair to a cheaper trusted local layer when possible, but retain the checker as the final semantic gate for high-risk sends.

We study a truthful facility location problem in which agents have private positions in a metric space and known approval preferences over two distinct facilities. Each agent incurs a cost equal to her total distance from the facilities she approves. The goal is to locate the facilities at a given set of candidate locations so as to approximately minimize either the social cost or the max cost, while incentivizing the agents to report their positions truthfully. This problem was previously studied for line metrics and deterministic mechanisms. We first settle the approximation ratio of deterministic mechanisms for the social cost on the line by establishing a lower bound of $7$, matching the upper bound from previous work, and show that randomization leads to a strict improvement. For the max cost, we show that the previously known optimal deterministic guarantee of $3$ remains tight even for universally strategyproof mechanisms. We then turn to general metric spaces. For deterministic strategyproof mechanisms, we obtain tight approximation ratios of $2n-1$ for the social cost and $5$ for the max cost. We also show that randomization yields substantial improvements for both objectives: for the social cost, we obtain a tight characterization, while for the max cost we establish nearly matching upper and lower

Surface reconstruction from range scans is challenging in the presence of noise and outliers, as fidelity to observations must be balanced against sensitivity to measurement errors. Yet, most neural approaches discard the sensor positions and corresponding visibility information, compensating for this loss with geometric priors or regularization terms. We instead formulate surface reconstruction as robust regression of the predicted first surface intersection along each acquisition ray against its measured depth, which penalizes predicted surface intersections within the segment between the sensor and the observed point. We represent the surface as the zero set of a 1-Lipschitz multi-layer perceptron, allowing the use of linearly convergent sphere tracing to compute the first hit, and use a robust Huber loss for regression. Our formulation requires neither oriented normals, sampled off-surface points, nor an eikonal loss. Comparisons with the state of the art on synthetic and real data with varying levels of noise and outlier contamination demonstrate competitive performance on clean data and much improved fidelity and robustness under even moderate corruption. Our analysis further confirms a trade-off between fidelity and robustness: greater fitting capacity improves data fidelity in the clean case but increases sensitivity to the measurement errors inherent in real acquisitions, with a sweet spot typically emerging at intermediate capacity.

Coding agents use generated tests to decide whether a proposed patch satisfies an issue's requirements. A suite that fails on the buggy program and passes after applying the reference patch can still accept an incorrect patch or reject a correct patch. We introduce SWEPATCHTEST, a test-generation benchmark with 619 tasks from SWE-bench Pro, SWE-bench Multilingual, and DeepSWE and 82,579 candidate patches collected from coding-agent trajectories. By replaying each suite on the buggy and patched codebases, \benchmark jointly measures issue reproduction, acceptance of correct patches, and rejection of incorrect patches. Joint Validation Rate (JVR), which requires all three criteria on the same task, exposes failures hidden by single-reference evaluation: the best-performing configuration achieves only 9.21% JVR despite reproducing the reference transition on 59.29% of tasks. We further use validation scores to select and scale demonstrations for supervised fine-tuning (SFT). The resulting training recipe approximately triples JVR and improves downstream repair across three SWE-bench benchmarks. Patch discrimination thus provides both a diagnostic measure of test quality and an execution-based signal for curating agent training data.

Compromised UAV reporters can poison latency telemetry and attract offloaded tasks to degraded servers. We present FleetGuard, an identity-capped conformal offloader for local, edge, and cloud execution. With equal buffers of m scores and a cap of K compromised identities, a Km rank correction retains one-sided fixed-level coverage without an honest-reporter majority. A deterministic repair scheduler allocates task-specific risk under fleet, capacity, and deadline constraints. Heavy-tailed simulations with eight reporters, four edges, 30 independent seeds, and 20 batches per seed and condition yield 2.08% deadline misses under seven low-score compromised reporters (95% clustered interval: 1.94-2.24%), compared with approximately 9.6% for pooled conformal, reporter-median, and mean rules. FleetGuard saves 22.3% energy over fixed-risk BRC and approaches adaptive exact assignment within 1.1%, with 2.6x lower mean scheduling time.

Diffusion language models (DLMs) enable flexible parallel generation by resolving tokens according to the model's evolving predictability rather than a fixed left-to-right order. This flexibility, however, introduces training choices absent from standard autoregressive teacher forcing: which reference tokens are visible when a target is supervised and how much generation space is allocated. Standard supervised fine-tuning (SFT) typically makes these choices through random masking, independently of inference. Our analysis shows how this training-inference mismatch can misdirect supervision. Random masking can conceal prediction errors or penalize uncertainty that decoding would otherwise resolve, while the attention-visible input length can change which response formulations the model favors, even when the underlying solution is unchanged. Motivated by these findings, we propose Predictability-Aware Canvas Selection and Supervision (PACS), an SFT method. PACS uses the model's own predictability to select each example's attention-visible input length and supervises reference tokens immediately before their confidence-guided revelation. Across reasoning and code-generation benchmarks, PACS consistently achieves the highest mean accuracy among the compared SFT methods, with gains of up to 12.84 percentage points over vanilla SFT on individual benchmarks. Code will be released upon publication.

Humanoid behavior foundation models (BFMs) enable diverse whole-body control, but adapting a pretrained controller to a new embodiment remains challenging because of kinematic and dynamic mismatch. We present TRACE, a framework for efficient cross-embodiment adaptation that preserves pretrained motion representations while specializing feedback control to the target robot.TRACE first injects kinematic priors into input and output projection layers and aligns nominal joint dynamics to reduce embodiment mismatch and support stable fine-tuning. It then combines encoder distillation with low-rank reinforcement learning on a compact, quality-screened motion set, using a two-phase curriculum to establish stable tracking before refining difficult motions. We apply TRACE to transfer SONIC from Unitree G1 to Unitree H2 using 10 million motion frames and approximately eight hours of training on a single RTX PRO 6000 GPU, at less than 0.1% of the GPU-hour cost of training SONIC from scratch. The adapted controller achieves 96.08% success on 67,953 evaluation motions under a height-scaled tracking criterion. Real-robot demonstrations span dynamic motion tracking and whole-body teleoperation, with collected demonstrations supporting downstream vision-language-action policy post-training. These results demonstrate practical reuse of a pretrained humanoid controller across robots with different body scales and dynamics.

Spiking Neural Networks (SNNs) are promising for energy-efficient neuromorphic computing, but they still suffer from catastrophic forgetting when learning tasks sequentially. This work investigates the integration of Experience Packing and Replay (EPR) as a memory-efficient continual learning strategy for deep SNNs under low-latency constraints. The proposed framework combines Batch Normalization Through Time stabilized VGG9 SNNs with Spike Activation Maps (SAM), which provide forward-only explanations based on spike activity. These explanations are temporally aggregated into spatial attention maps and used to select the most informative image patches for replay, instead of storing full images. Experiments on Split CIFAR10 and Split CIFAR100 show that SAM-guided EPR improves the accuracy-memory trade-off under tight memory budgets, and remains competitive while requiring substantially less storage.

Research evaluation has developed around quantity, citation impact and, more recently, the accumulation of research excellence, but these dimensions remain largely dependent on citations, which describe how research is received rather than what it contains. We propose uniqueness as a complementary dimension and measure it with the Research Uniqueness Index (RUI), which asks how far a university's research sits from what is typical in its field, using no citations but the content of the research itself. Each researcher is represented by an embedding of their abstracts, and an unbalanced version of Optimal Transport returns the share of that work which cannot be matched into a random sample of same-field researchers, the standard population; that unmatched share is the score. Because every researcher is scored, a university is a distribution rather than one number, and we treat that distribution as the indicator: it shows whether uniqueness is spread across an institution or concentrated in a few groups. Quantiles summarise it and, being quantiles, do not reward size. Since uniqueness is defined only relative to a field, we demonstrate the index within one, neuroscience (Scopus, 2021-2025; 9,861 reference researchers). All ten universities studied sit well above the standard (RUI_median 0.40-0.53 versus 0.27) and five specialised institutes higher still (0.57-0.73), the most unique work often confined to a narrow niche that the same scores name. RUI is uncorrelated with citation impact, and a controlled test confirms that it tracks displacement from the field rather than topical concentration; results are robust to an alternative embedding and replicate in computer science and economics. The index is not intended as a league table: the gap above the standard and the researcher-level niche structure are stable, whereas the order among adjacent universities is not.

Sugarcane leaf diseases severely impact crop yield and quality, making accurate and timely detection critical for sustainable agriculture. Conventional deep learning models for plant disease detection depend on large annotated datasets, which are expensive to obtain and challenging to prepare due to challenges commonly observed across crop disease detection tasks, including (i) high intra-class variation caused by diverse symptom expressions across growth stages, (ii) subtle inter-class similarities between diseased and healthy tissues, and (iii) the risk of noisy labels degrading model reliability. To handle these challenges, we presented an Uncertainty-aware Mean-Teacher based Semi-Supervised Learning Disease Detection Network (U-SSDNet) that leverages unlabeled data while maintaining robustness. First, discrete wavelet transform preprocessing is applied to enhance lesion visibility and reduce intra- class variation. Subsequently, the KL divergence is employed to refine the pseudo-labels and reduce the impact of confirmation bias. At the same time, supervised contrastive learning is applied to make the features more distin- guishable and improve the separation among different classes. Experimental evaluations on sugarcane leaf disease datasets, considered a representative case study, demonstrate that the proposed approach achieves performance close to that of fully supervised models while requiring far fewer annotated samples and consistently outperforms state-of-the-art supervised and semi-supervised approaches.

Symbolic regression is the problem of finding an expression that fits a dataset. Solving it requires searching a combinatorial space of candidate expressions. Neural symbolic regression can amortize that search by training a model once on synthetic data and then predicting expressions for a new dataset. But existing models still underperform genetic programming on real, noisy data. We argue that the deficit lies in the training rather than in the paradigm, and present RASR (Reinforced Amortized Symbolic Regression), a method with two key changes to the training recipe. First, we design a synthetic prior that narrows the gap with real-world data: inputs come from the prior of a tabular foundation model; constants are sampled bottom-up, conditioned on the data; targets carry heteroscedastic and structured noise; and everything is standardized. Second, we split training into two stages. Supervised pre-training maximizes the token-level likelihood of a reference expression. This is followed by reinforcement-learning post-training that aligns the model with its inference-time use by rewarding held-out fit and penalizing expression size. On the black-box track of SRBench 2.0, RASR achieves the best mean test $R^2$ rank of all evaluated methods. Three RASR variants, trained with different size penalties, make up most of the accuracy-complexity Pareto front. On Strogatz, our method recovers more ground-truth expressions than any baseline, and on Feynman it is competitive with the best-performing methods.

Certified machine unlearning ensures the proximity between the unlearned model and a model retrained without the data to be forgotten. Extending this framework to sequential deletion requests, typically arriving as groups of variable size (e.g., all records of one user or of coordinated users), is challenging. For unlearning methods based on fine-tuning on the retain sets, this leads to either degrading performance or increasing computation as the number of unlearning requests grows. For independent successive requests, we instead certify each model release against a counterfactual unlearning trajectory in which the group being deleted never took part. With a fixed fine-tuning budget per deletion, we show that, for smooth strongly convex objectives, this approach keeps the performance loss bounded, uniformly in the number of requests. Experiments with multiclass logistic regression on FEMNIST show that the resulting algorithm needs fewer corrective steps than existing sequential methods and outperforms training-time differential privacy, particularly when group sizes are heterogeneous.

Recovering high-quality images from degraded underwater observations is crucial for underwater exploration. Polarization observations show promise as an effective source of complementary physical cues for recovering target radiance under clear-water conditions. However, existing methods have yet to jointly model raw intensity information and polarization cues to enable image restoration driven by an understanding of mixed degradations (color casts, haze, and low light). To overcome these limitations, we propose PolUR, a unified multimodal language-driven framework that jointly uses polarization and RGB information to identify degradation types and severity levels through structured semantic instructions and enhances joint restoration representations under the guidance of degradation-aware semantic features to improve restoration robustness under mixed degradations. We further construct a new benchmark named MixPolar, which contains underwater images exhibiting real-world degradations and their corresponding degradation descriptions. MixPolar supports systematic evaluation of degradation perception, image restoration, and performance on downstream tasks. Extensive experiments demonstrate that PolUR effectively integrates polarization and RGB information, achieving competitive performance in degradation perception and image restoration, while also demonstrating generalization capability and utility for downstream tasks.

A medical vision-language model may analyze a given CT scan and correctly detect both kidneys in view. Continuing this conversation, if a user provides a prescription with an opposite direction, medical VLMs often sycophantically choose to realign their diagnosis. In this work, we study regressive sycophancy under fabricated evidence, where a model abandons a correct answer when pushback comes with a fake document. We test four open VLMs on 400 VQA-RAD yes/no questions in 19,200 multi-turn conversations, with and without a fabricated prescription. Without a disclosure that documents are synthetic, the models abandon 80.1% of 6,488 correct answers. Notably, under one script each, the same prescription raises the flip rate of Qwen2.5-VL-7B from 45.9% to 89.6% and lowers that of Qwen3-VL-8B from 77.7% to 50.9%. A model rejected the prescription while still preserving the correct answer in 915 of 3,232 conversations with the prescription (28.3%), and 40% of these flipped in later turns. System-prompt sentences that disclose the documents as synthetic lower one model's flip rate from 77.8% to 47.3%. Thus, we conclude that awareness about a forgery does not prevent a model from adopting its answer.

A common practice in instance retrieval is to grow the gallery size to improve a benchmark's discrimination capability. One prominent example is ILIAS, which adds up to 100 million distractor images. The prevailing assumption is that faithful evaluation demands web scale, and that indexing all 100M images through the backbone is the unavoidable price of a comparable number. That price is steep: hundreds of GPU-hours and ~200 GB of descriptors per model, paid again for every method or ablation. We show the web-scale indexing price is largely avoidable: evaluation itself obeys a scaling law. The number of distractors that outrank a positive is Binomial in the gallery size. We define the per-query retrieval hardness, a gallery-scale agnostic quantity that yields a retrieval metric at any scale, and demonstrate it can be derived from a small random sample. The underlying sampled-rank machinery is long established in recommender-system and biometric evaluation, but to our knowledge has never been adapted to visual retrieval. We make the law per-query and estimable, and validate it at true 100M scale on ILIAS and INSTRE, and on the full 759k-distractor Google Landmarks v2 pool. Concretely, within a 5% sample (a flat 20x less indexing), every backbone's tail-aware (EVT) estimate has converged onto the large-scale mAP, to within ~0.01 (mean absolute error 0.005), across five backbones, three benchmarks, and two distractor pools (mAP@1000 at 100M on ILIAS/INSTRE; the official mAP@100 at the full 759k pool on GLDv2); the recommender-systems count baseline over-predicts everywhere (+0.037 to +0.068). This is an order of magnitude better than naively reporting the metric on the same subset (+0.08 to +0.13). We release code and the per-query hardness to facilitate faster, cheaper, and greener retrieval evaluation.

Recent Vision-Language-Action (VLA) models, such as $\pi_{0.5}$, have demonstrated impressive generalization capabilities by scaling behavior cloning across massive datasets. However, these models frequently suffer from distributional convergence, a phenomenon where the policy regresses to "safe" but suboptimal mean behaviors due to mode averaging in the training data. This often results in generic grasp strategies, Convergent Defaults, that fail in constrained or long-tail scenarios. In this work, we propose a novel framework to identify and correct these behaviors without parameter updates. We introduce a three-stage methodology: (1) Probing for Defaults, identifying task-specific mode collapse; (2) Diagnostic Root Cause Analysis, utilizing the VLA's latent reasoning capabilities to verbalize the logic behind suboptimal actions; and (3) Structured Guidance Injection, where concrete, distinct alternatives are encoded into reusable "Lens Prompts" that shift the sampling distribution. We validate this approach using $\pi_{0.5}$ in ManiSkill3 simulation environments. We hypothesize that our method outperforms vanilla $\pi_{0.5}$ and standard Chain-of-Thought prompting by effectively breaking convergent defaults in complex manipulation tasks.

Plurality-voting ensembles aggregate the predictions of multiple base classifiers and output the most voted class, with ties typically broken at random. Despite the simple aggregation rule, attributing the impact of each base classifier to the ensemble's predictions is far from straightforward. That is, a single base classifier's impact depends on how it interacts with the votes of all other classifiers in the ensemble, for example, in resolving ties, and not only on its own predictions. We address the attribution problem using established tools from cooperative game theory, studying the Banzhaf and Shapley power indices for new games modeling plurality voting and random tie-breaking, for binary and multiclass classification settings. Computing power indices is, in general, intractable. Surprisingly, our methods, collectively named ENPOWER, enable the exact computation of Banzhaf and Shapley values in linear time for the binary classification setting and in low-degree polynomial time for the multiclass setting. The algorithms in ENPOWER rely on new combinatorial formulations of the Banzhaf and Shapley indices, yielding closed-form solutions for the binary case and ordinary generating function representations for the multiclass case. We validate empirically the methods in ENPOWER demonstrating their high applicability, providing attribution values not captured by existing methods. We apply the attribution of ENPOWER to: ensemble pruning and prompt-based ensembling of vision-language models.

Encrypted traffic classification is critical for network management and security, but practical deployment must balance classification accuracy and inference cost. Existing models typically process fixed flow windows with limited control over computation after training, which can limit the accuracy-efficiency trade-off under constrained resources. To address this problem, we present NetCoFi, a budget-aware classifier that allocates computation at coarse and fine granularities. Using byte representations and observable packet metadata, NetCoFi selects token neighborhoods for fine-grained encoding while compressing the full window into complementary coarse context. Multi-budget training enables the same model to operate at different routing budgets without retraining. Evaluation on six public datasets shows that NetCoFi achieves state-of-the-art classification performance when trained from scratch. Under low-budget settings, NetCoFi improves average Macro-F1 by 7.86 percentage points across three datasets, compared with a same-depth static baseline. In a DPDK-GPU traffic processing pipeline, it increases mean batch throughput from 26.42 to 43.42 Mb/s, a 1.64x improvement over NetMamba.

Long-horizon robotic manipulation requires robust execution over sequences of interdependent actions. Existing methods typically handle execution deviations by replanning the remaining action chunk or correcting actions step by step as new observations arrive. However, a local deviation may affect only a few actions under distribution shift, yet replanning regenerates the entire chunk from an unfamiliar state, potentially introducing errors into otherwise valid actions. Step-wise correction may react too late, as it does not explicitly anticipate how the deviation affects subsequent actions. We propose ReEnvision, a future-guided selective action repair framework that updates a joint future-action belief using execution feedback. It then learns how the resulting belief revision affects each unexecuted action. The resulting belief revision enables selective repair of only the affected actions within a cached action chunk, preserving valid actions and reserving full replanning for severe deviations. On CALVIN, L-CALVIN, and RLBench, ReEnvision achieves 86.6% 5-step success, improves 10-step success from 56% to 63%, and reaches an average success rate of 88.8%, respectively. On five real-world manipulation tasks, it achieves an average success rate of 82.8%. Controlled repair experiments further demonstrate the benefits of localizing affected actions and preserving valid ones during execution.

Do vision-language models use depth information to judge which object is closer, or rely on image-plane cues? We investigate this question through controlled depth–cue conflicts, representation analysis, and interventions. Among the many image-plane cues relevant to depth perception, we focus on apparent size and vertical position. We compare depth judgments under cue-consistent and cue-conflict conditions in both rendered scenes and real-world photographs. Across model families and scales, depth judgments systematically shift toward misleading cues. Yet the correct depth order remains decodable from the queried objects' visual representations, including when the model answers incorrectly. Image and category controls support interpreting this signal as image-dependent depth information beyond the two target cues. Strengthening the model's own depth readout through object-level write-back can correct conflict decisions without updating model parameters or using ground-truth depth annotations. Interventions show that the effect of this correction depends on the direction and target tokens of the written signal, while its strength varies across architectures and domains. Together, these findings reveal a mismatch between a model's ability to represent depth and use this information when deciding the answer. VLMs can retain useful depth information while allowing competing image-plane cues to exert greater influence on their answers.

We introduce STRUCTURES-P, a machine-learned all-electron orbital-free density functional for periodic CHNO systems. Electrostatics are evaluated analytically in an atom-centered Gaussian basis, while an equivariant network learns the kinetic and exchange-correlation energy from energies and gradients of non-ground-state densities. Minimizing the functional from a superposition of atomic densities recovers the Kohn-Sham ground state: on held-out crystals, the density error falls from 6.7% to 0.4% of the electron count and the energy error reaches 0.25 mHa per atom, half that of a same-architecture interatomic potential. A single functional can be trained for both isolated molecules and periodic systems, and the model generalizes to larger unit cells. It largely preserves relative-energy rankings and reproduces convex-hull vertices on the tested H-N composition slice. Together, these results demonstrate the viability of machine-learned all-electron orbital-free DFT for variational density optimization and energy evaluation in fixed-geometry solids.

Immersed-boundary methods avoid body-fitted meshes, but the energy or entropy stability of the underlying flow discretisation does not automatically extend to the immersed treatment. This manuscript analyses the volume-penalised compressible Navier–Stokes equations for both energy and entropy stability, on the continuous and semi-discrete levels, and integrates the isolated source exactly in time. The source is formulated as a one-parameter family indexed by $\theta$, whose momentum component relaxes the fluid velocity towards the prescribed solid velocity; $\theta$ sets how much of the resulting kinetic-energy loss leaves the domain rather than being converted to internal energy. The two stability requirements impose opposing bounds on $\theta$, with dissipation of the relative total energy requiring $\theta\geq0$ and stability of the mathematical entropy requiring $\theta\leq1$. Where penalisation acts, both conditions are therefore satisfied if and only if $0\leq\theta\leq1$, and the symmetric value $\theta=1/2$ divides the kinetic-energy loss equally between the domain and the internal energy. We conclude that any entropy-stable split-form discontinuous Galerkin spectral element method remains semi-discretely entropy stable once the source is added at the collocation nodes, for stationary and prescribed moving bodies alike.

The semantic entropy of a language model is the entropy of the meanings of its sampled answers. The meanings are never observed, and the entropy is estimated from the verdicts of a pairwise equivalence judge. Given the meanings of $n$ answers, a known channel declares each pair equivalent independently, with probability $q+\Delta$ within a meaning and $q$ across meanings. At most $K$ meanings occur, and $q$ is bounded away from $0$ and $1$. For a weak judge, with $n\Delta^2\to0$, the global minimax rate is set by distributions near a vertex of the simplex and overstates the difficulty elsewhere. We show that the difficulty is set by how many meanings can be rare. Over distributions with at most $s$ meanings below a fixed probability $\tau<1/(K-s)$, the minimax squared risk is of order $N^{-1/s}$ for $s\ge2$, $N^{-1}\log^2(e+N)$ for $s=1$ and $N^{-1}$ for $s=0$, capped at one, where $N=n^K\Delta^{2K-2}$. The matching lower bounds hold on every fixed polynomial window $n\Delta^2\le n^{-\kappa}$. One estimator attains these rates without knowing $s$ or $\tau$: it averages products of centered verdicts along all spanning trees of every $k$-set of answers, projects the resulting power sums onto the valid moment set and inverts them. The case $s=K-1$ gives the global rate $N^{-1/(K-1)}$ and the consistency threshold: with $\Delta=n^{-\gamma}$ the entropy is uniformly consistently estimable exactly when $\gamma<K/(2K-2)$. On polynomial windows, the same estimator attains the sharp local minimax constant on shrinking neighborhoods of interior points with distinct probabilities. Restricting the average to stars multiplies the constant by $K^{K-3}$.

Long-running agents retain KV caches across turns and sessions for hours or days. Compressing these persistent caches requires reducing storage while controlling reconstruction error. We present AlignPred, a codec that combines local alignment with predictive coding under a configurable KV reconstruction-error budget. Local orthogonal transforms align related heads or layers into a common coordinate frame, allowing AlignPred to combine token-wise prediction with same-token prediction across heads or layers. Residuals are quantized on an error-bounded lattice and entropy-coded, without learned predictors, encoder-side mode search, or large joint transforms. Across three attention mechanisms, AlignPred requires less storage than the evaluated baselines at comparable measured reconstruction error. Against the evaluated CPU codec implementations, it decodes at least 1.6 times faster while matching or improving encode time. AlignPred achieves near-native task scores at substantially compressed operating points on LongBench-Pro at 128K context. AlignPred's SGLang HiCache integration reduces persistent target-KV storage by 42-72% relative to native HiCache, with comparable latency.

Modern Transformers rely on many scale-control mechanisms, including normalization, residual scaling, attention temperatures, loss weighting, and adaptive updates. To explain how these mechanisms maintain scale consistency, we introduce Neural Dimensional Analysis (NDA), a unified framework based on dimensional consistency. We define neural units that specify how tensors rescale and which quantities can be added, just as physical units do in physics. Our rules match residual units, nondimensionalize attention and gate inputs, combine losses across units, and convert gradient units into parameter units. We implement NDA through dimensionless Unit Violation Indices (UVI) for diagnosing scale mismatches within Transformers, a unit-consistent multi-loss objective, and the Gauge-Canonical Transformer (GCT) with normalized coordinates and orthogonalized residual additions. In language-model experiments spanning 51M to 1B parameters, GCT remains stable during training across a broad sixteenfold learning-rate range, outperforming Pre-LN, LLaMA-style, and QK-Norm in 124M validation loss at tuned learning rates. GCT's normalized residual stream also shows improved robustness to quantization. At 1B parameters, calibrated per-tensor int8 quantization increases GCT's validation loss by only 0.009 nats, compared with up to 5.51 nats for the baselines. Vision and physics-informed neural network experiments further confirm NDA's benefits beyond language modeling. These results establish dimensional consistency as a neural network design principle.

Diffusion MRI (dMRI) probes water diffusion within the brain, providing unique information about tissue microstructure and white matter connectivity. High-quality dMRI enables advanced and accurate analyses but demands dense acquisitions, resulting in long scan times and hardware constraints that are often infeasible. Further, dMRI requires dedicated computational methods to support diverse analysis tasks (e.g., parcellation, registration, and synthesis) for both neuroscientific and clinical applications. Here, we present dMRI-F, a foundation model that unifies dMRI data enhancement and downstream tasks across diverse acquisitions and populations, while requiring only minimal acquisition input. dMRI-F is pre-trained with a q-space masked reconstruction task: predicting densely sampled multi-shell data from only six volumes. It learns generalizable latent feature representations of dMRI data, which serve as a core foundation for adapting to diverse downstream tasks, including brain parcellation, brain registration, and cross-modal synthesis. Across nine independently acquired datasets, we demonstrate that dMRI-F outperforms state-of-the-art methods with robust generalization across acquisitions, populations, and tasks. Our results demonstrate dMRI-F's versatility as a general-purpose framework that enables diverse dMRI applications from low-cost, sparse acquisitions.

Long-context inference makes the key-value (KV) cache a major memory bottleneck for large language models. We propose BELT, a low-rank KV cache compression framework. Its key-side objective targets attention distributions rather than raw keys or pre-softmax logits, yielding a quadratic surrogate with a closed-form low-rank solution. A learned global rank allocator uses differentiable soft selection during training and hard selection at deployment to meet a user-specified component budget. Experiments demonstrate improved model quality under aggressive compression and increased decoding throughput at comparable accuracy. Evaluated across three LLMs and multiple benchmarks, BELT achieves up to 70% low-rank KV cache compression while outperforming the strongest baseline by up to 23.3 points on LongBench, and up to 17.8x overall KV cache reduction when combined with quantization. With Triton-based GPU kernels and a quantized cache, BELT delivers up to 2.7x speedup for the attention module and 3.3x end-to-end decoding throughput.

This paper shows how a Schrodinger bridge (SB) structure emerges from maximum-entropy (MaxEnt) distribution steering of discrete-time nonlinear systems. We also establish that the resulting bridge can be realized by an optimal control policy. We consider general state and input costs, non-Gaussian endpoint distributions, and input constraints. Assuming the existence of a feasible policy with finite objective value, we show that the state-path measure under a MaxEnt optimal control policy is the unique SB. The reference measure of this SB is induced by an optimal policy of the corresponding terminal-unconstrained problem. Moreover, we give an equivalent characterization of the optimal state-path measure using a killed reference and its normalized Feynman–Kac measure, thereby interpreting the state cost as killing. We then characterize the finite-cost feasibility by the existence of an endpoint coupling having finite Kullback–Leibler divergence from the reference endpoint distribution. For compactly supported marginals with bounded target density, we derive reachability-based sufficient conditions guaranteeing this feasibility. We further provide specializations to fully actuated control-affine and reachable linear systems.

Brick-and-mortar retailers face many different challenges that involve understanding thoroughly its products catalog and customer preferences. In particular, assortment optimization - proposing the ideal mix of products - and promotion planning hold a pivotal role in their strategy. By leveraging sales data, retailers can make informed decisions on which products to sell and how to manage inventory, based on customer preferences as well as regional and seasonal trends. It is especially crucial to capture interactions between products, in order to minimize the number of items that cannibalize each other's sales and to ensure that complementary products, which are often purchased together, are conjointly available and sold across all stores. In this paper, we propose a model of shopping basket that learns embeddings to represent interactions between products, prices and stores. Our model is built to uncover sales patterns from very large transaction datasets. In particular, the optimization loss is computed with random negative samples in order to overcome the computational bottlenecks that arise with large number of items. Our experiments on synthetic data show the efficiency of drawing such negative samples based on the actual assortment of available products, with better results than approaches from the literature. We also validate our approach by training and evaluating our model on a dataset composed of millions of transactions from Fortenova Group, one of the leading retail company in Southeast Europe. Our model showcases promising applications in the sector of retail, with enriched interfaces to efficiently support category managers.