A recent paper of Barron et. al. (2026) identified two numerical issues with the softmax and cross-entropy functions in virtually all standard machine learning libraries (e.g., PyTorch and JAX). We show that these issues can be recovered through simple property-based tests in Hypothesis. However, our tests identify an additional edge case where their implementation fails. We then propose an even more robust softmax implementation which passes these tests and also supports use in causal attention masking. We compare our implementation with theirs and discuss some of the differences. Overall, this work shows that property-based testing can be a powerful tool for discovering and fixing numerical issues in machine learning libraries. Our implementations and property-based tests can be found at https://github.com/mmaaz-git/softmax-fix.
The goal of Quality-Diversity Reinforcement Learning (QD-RL) is to find a population of high-performing policies that solve a decision-making problem in diverse ways. Despite recent progress, QD-RL methods predominantly rely on evolutionary operators--including black-box optimizers, mutations, and crossovers--to facilitate exploration and induce diversity. Scaling these methods becomes increasingly costly as policies and behavior spaces grow larger. Building on the Soft QD view of quality-diversity optimization, we introduce \textbf{Re}pulsive \textbf{P}opulation \textbf{P}olicy \textbf{O}ptimization (\textbf{RePPO}), which adapts PPO to jointly optimize a compact population of diverse, high-quality policies using a population-dependent, dynamically shaped reward. On standard continuous-control tasks, RePPO achieves a strong quality-diversity tradeoff compared with state-of-the-art QD-RL methods. Ablations further characterize this tradeoff by examining how each component affects individual quality and collective diversity. To assess RePPO's scalability, we extend our evaluation to learning from pixels and fine-tuning large language models. In pixel-based tasks, where large policies make most traditional QD-RL methods infeasible, RePPO maintains quality on par with standard PPO while achieving substantially higher diversity. Its successful application to language model fine-tuning further demonstrates, for the first time, that QD-RL methods can scale to this domain. Together, these results show that directly optimizing compact populations with RePPO provides a scalable approach to QD-RL across increasingly complex domains.
In this paper, we compare the performance of three popular deep learning architectures: Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and CNN-LSTM in water level prediction over different time window sizes. We used wind, pressure, and tide data from National Oceanic and Atmospheric Administration (NOAA) stations. Our results show that the LSTM model generally outperformed the other models in forecasting water level trends. By applying LSTM, we achieved a mean squared error (MSE) of 0.006, making it the most reliable option for real-world use. We predicted water levels using wind, pressure, and tides. Wind, pressure, and tide data are time-based features that lack spatial patterns. Consequently, LSTM performed better than spatial methods, such as CNN. We also investigated Hurricane Ian, which occurred in Tampa Bay in September 2022 and caused a significant storm surge. Our findings show that CNN and CNN-LSTM performed slightly better than LSTMs during powerful storms, such as hurricanes. Hurricanes are seldom observed in training data, and CNN-based models are more reliable when predicting data that were not included in the training. To explain how environmental factors affect the predicted water level, we used the perturbation method. The perturbation shows that the tide is more effective at predicting water levels than wind and pressure. This study demonstrates the potential of deep learning to improve storm surge prediction accuracy.
Attack logs and red-team reports show targets the attacks used against them, to help them repair their defenses. Prompt injections against large language models (LLMs) are plain text, so such records also hand targets a working copy. Drawing on social learning theory and work on the victim-offender overlap, we study what targets do with that copy in Tensor Trust, an online game whose players defended LLM-guarded accounts, attacked one another's, and could read every attack they received. Using all 563,349 attack records, we compare each received attack with matched attacks that circulated among similar players at the same moment but never reached the target. Reuse mainly followed success: within a week, targets attacked other players with 10.0% of the attacks that had broken their defense, versus 0.03% of matched attacks (adjusted difference: 8.8-10.0 percentage points). Near-misses, whose replies began with "access granted" but that scored as failures, were reused about as rarely as failures: reuse tracked whether an attack paid off, not how its reply read, consistent with payoff-biased social learning. Breaches were also followed by defense changes, after which same-family attacks became 24.5 percentage points less successful against the target, relative to other defenders. Reuse came mainly from players already attacking, not newcomers. Previously received texts accounted for 23.2% of the successful hijacks we analyze. The attack history thus served both repair and reuse; we derive testable hypotheses for how AI security tools should disclose attacks, such as describing the technique without the full text.
Valentin Leplat, Daniil Merkulov, Aleksandr Katrutsa, Daniel Bershatsky, Olga Tsymboi, Ivan Oseledets
原文摘要
Stochastic first-order methods are widely used for large-scale optimization and machine learning, where their performance often depends on the learning rate. We study NAG-GS, a stochastic momentum method based on a semi-implicit Gauss--Seidel discretization of a Nesterov-type flow. For strongly convex quadratics, we determine the exact range of stable step sizes for fixed damping and the step size that minimizes the worst-case spectral radius. When the damping equals the strong-convexity parameter, the method recovers the accelerated asymptotic factor $(\sqrtκ-1)/(\sqrtκ+1)$, where $κ$ is the condition number. We also prove that the quadratic iteration converges linearly with dynamic damping when its limiting fixed-damping iteration is stable. For general strongly convex, possibly nonsmooth objectives, we establish an $O(1/k)$ bound on the expected objective gap of the last iterate, using a conditionally unbiased stochastic first-order oracle with bounded second moment. The result holds with dynamic damping from any $γ_0>0$. We show that NAG-GS can also be written as a Heavy-Ball iteration with explicitly determined step sizes and momentum coefficients. We test the stochastic result on nonsmooth hinge-loss classification problems, and further evaluate NAG-GS on quadratic problems, logistic regression, convolutional neural networks, RoBERTa, and Vision Transformers. The experiments show that NAG-GS can remain effective at learning rates where standard optimizers become unstable or lose accuracy.
Low-cost arms such as the SO-101 have become the most common entry point to robot learning, yet no released policy performs spatially constrained, language-conditioned tasks on them out of the box, and existing SO-101 benchmarks run only on physical hardware with simple tasks and few held-out objects. We introduce SO-101 Bench, a simulated benchmark for language-conditioned spatial manipulation built on an Isaac Lab digital twin of a real SO-101 setup. It has four tasks scored as five families, 138 objects in seen, unseen-instance and unseen-class splits, 916 evaluation episodes with subsets for held-out appearances, rotate-to-fit planning and occlusion memory, and 3,629 simulated demonstrations on which every learned policy is fine-tuned under an equal compute budget. The twin ranks policies as the real robot does: five policies trained without real data, four VLAs and an LLM agent, keep their exact ranking when the same weights run on the physical arm in an unseen environment with unseen objects (Spearman rho = 1.00, MMRV = 0.000), although simulation overestimates success. Across nine VLAs and world-action models, the best learned policies score near 61%, with spatial placement, unseen object classes and rotate-to-fit planning remaining open. An LLM agent that writes code and hands grasping to a learned policy scores 83.2%, and an ablation shows that this lead comes from the choice of harness and the choice of language model.
Jordan L. Cahoon, Chloe Stanwyck, Asad Aali, Rachel Madding, Sulaiman S. Somani, Emma Sun, Yixing Jiang, Renumathy Dhanasekaran 等 9 位作者
原文摘要
Background: Clinical notes contain extensive duplicated text from templates, copy-paste, and auto-populated fields ("note bloat"), diluting clinical signal, limiting longitudinal context, and increasing large language model (LLM) costs. Methods: TRACE removes note bloat using note-level EHR metadata to identify templated and copied content, with frequency-based de-duplication when metadata are unavailable. We evaluated TRACE using blinded physician span review and gold-standard templated-text annotations across four cohorts spanning liver transplant, obstetrics, and inpatient populations at multiple health systems (5.3M notes). We compared zero-shot LLMs and embedding-based classifiers using original and TRACE-processed notes for 20 information extraction tasks and prediction of 5-year survival, postpartum hemorrhage, and 30-day readmission. Results: Only 0.3-6.6% of removed text was flagged as author-generated; TRACE captured 86% of annotated templated characters. Information extraction F1 differences averaged by cohort ranged from -0.009 to +0.004; task-specific prediction F1 differences ranged from -0.011 to +0.018. Among 1,000 randomly sampled Stanford Health Care patients, TRACE reduced chart text by 47.3% (742.7M characters), averaging 220,167 fewer tokens per patient. Using 2024 encounter volumes at a large tertiary academic center and one query per encounter, projected three-year net savings ranged from $1.00M to $13.58M across evaluated model pricing schemes, including initial and annual TRACE processing costs. Conclusion: TRACE substantially reduces clinical note redundancy while preserving information extraction and prediction performance. Underused EHR metadata can reduce LLM inference costs, expand usable longitudinal context, and support scalable clinical AI.
Bingliang Zhang, Wenda Chu, Yizhuo Li, Linjie Yang, Yisong Yue, Katherine L. Bouman, Yang Song, Qiushan Guo
原文摘要
We present Scalable Pixel-anchored End-to-end Diffusion (SpeeDiff), a latent diffusion method that jointly trains the VAE and the diffusion model from scratch. In principle, joint training allows the diffusion loss gradient to directly guide the VAE encoder, encouraging the formation of a generation-friendly latent space and potentially yielding faster convergence than the conventional two-stage approach with a pretrained frozen VAE. However, a naive end-to-end implementation severely degrades performance, as unrestricted backpropagation of the diffusion loss leads to latent space collapse. Our main technical contribution is a simple yet effective Tweedie Pixel Reconstruction (TPR) loss, which provides additional pixel-level feedback by decoding a predicted clean latent from an intermediate noisy state using Tweedie's formula, thereby alleviating collapse. Furthermore, our method enables jointly scaling a fully transformer-based architecture and enhances representation alignment within the end-to-end framework. Our SpeeDiff-XL model achieves over 140x and 61x faster training compared to Vanilla SiT and REPA, respectively, while attaining an FID of 1.50 without guidance on ImageNet 256x256 generation. With a more efficient 32x compressed VAE, our model further reaches an FID of 1.53 without guidance on ImageNet 512x512 generation.
Tool-assisted planning success can reflect external search and verification, leaving a model's unaided planning ability unresolved. We isolate innate planning through complete plans generated in one response without tools or execution feedback. We propose a verifiable framework that decomposes success-weighted plan efficiency as $\mathrm{SPE}=R\times Q\times E$: execution reliability, planning coordination among legal plans, and action economy among completed plans. Trajectory replay and certified optima verify these factors; the joint score includes unsuccessful attempts. We evaluate 8,400 full-plan outputs from seven model conditions on $4\times4$ sliding puzzles, controlling exact horizon and obstacle count in the primary cohort. Models that complete every trial still become less economical at longer horizons, while higher efficiency among completed plans can coexist with poor goal attainment. SPE separates the overall capability hierarchy, while its factors distinguish coordination failure from inefficient completion. The framework makes long-horizon limitations in unaided planning quantitatively verifiable.
RAG systems face a capability-cost gap: lightweight retrievers locate evidence but expose no operands for exact corpus-level operations (counts, intersections, aggregations), while knowledge ingestion systems obtain them only through expensive ingestion. ScalableRAG closes this gap with a persistent typed-set workspace: an agent builds named document and value sets at query time and composes them through set algebra, typed extraction, aggregation, and diagnostics. Zero-Ingestion ScalableRAG needs no preprocessing, not even a vector index. It matches the leading baseline on four conventional QA benchmarks (74.75 versus 74.64) and on aggregative Hotels-50 beats every evaluated system, scoring 71.74: 36.23 above the leading baseline (SRAG), and 40.22 points above A-RAG. Removing the workspace drops Hotels-50 to 38.4. On Hotels-350, the partition SRAG itself reports on, ScalableRAG leads SRAG by 9.3 points. Limited-Ingestion ScalableRAG adds a lightweight index and is the only evaluated system above 80 judge score at 5M documents, at 13 to 16 seconds per question and under 20 minutes ingestion; at 1M it scores 84.5 to A-RAG's 80.0 with 16x lower latency and 0.5% of its ingestion time. ScalableRAG combines minimal ingestion, conventional-QA parity, exact corpus operations, and million-document viability; no evaluated alternative provides all four. Our code is available at https://github.com/cohesity/ScalableRAG .
Chao-Yeh Chen, Tiffany Chen, Lin Duan, Tianyi Hu, Yu-Hsiang Huang, Umar Iqbal, Jiefeng Li, Edy Lim 等 15 位作者
原文摘要
We introduce EgoHand, a monocular 3D hand reconstruction model that achieves state-of-the-art performance on egocentric video. Central to our approach is the curation of a high-quality training corpus of 14 million images, combining synthetic data, in-the-wild video, and multi-camera motion capture to balance scene diversity and supervision quality. The model adopts SOMA-Hand as its parametric representation and processes 2D hand crops with a transformer encoder, incorporating camera cues and an auxiliary 2D branch to improve geometric consistency and image alignment. Zero-shot evaluations on HOT3D, ARCTIC, and H2O demonstrate strong out-of-domain generalization, with lower joint-position, root-translation, and acceleration errors than competing baselines at an inference latency of approximately 9.01 ms. A blinded user preference study further shows that participants favor EgoHand's reconstructed hand poses over those from Apple EgoDex and Ropedia Xperience-10M. We release the EgoHand model and source code under the OpenMDW license, allowing free commercial use of the model.
Generative AI can produce code, tests, documentation, and technical explanations that resemble artefacts traditionally used as evidence of student competence, weakening the inference from artefact quality to the engineering capability behind it. We report the redesign and first deployment of a 13-week, AI-permissive, software-intensive engineering course in which students may delegate substantial work to AI while retaining responsibility for planning, verification, review, integration, and explanation. The redesign couples an engineering accountability workflow, which structures human-AI work and preserves human decision authority, with an assessment evidence architecture comprising six complementary channels: product, process/origin, verification, explanation, reflection, and individual assurance. A collect--cross-check--escalate (CCE) reasoning logic connects these channels rather than treating any single artefact or trace as sufficient evidence of competence. By the end of Week 11, the deployment involved 87 students organised into 17 teams, with 957 individual weekly Git repositories and 34 group-project Git repositories supporting repeated assessment and engineering activity. The first deployment yielded four practical lessons: AI assistance changes where rubric criteria carry evidential weight; multiple evidence channels require deliberate cross-checking and meaningful complementarity; the evidential strength realised in practice may differ from that intended in the assessment design; and automation scales evidence collection more readily than interpretation. The contribution is a conceptually grounded, deployed design for making engineering responsibility and competence more visible when AI can perform substantial artefact production.
Temporal Predictive Coding (tPC) networks offer a layer-local alternative to backpropagation. For linear tPC, we show, by applying tools from cellular sheaf theory, that temporal credit assignment is a tug-of-war between spatial displacement and temporal errors, whose ratios are modulated by persistence and exposure operators $P$ and $Q$. Spatial inference therefore gives us the effective recurrent transition $F=PA$, whilst future credit is filtered by $P^\ast$. This creates a persistence--error tradeoff in spatial modes: under non-amplifying invariant recurrence, modes that preserve influence over long delays simultaneously fail to expose that influence strongly as temporal error. Recurrence can circumvent this via routing between persistent and exposed modes.
Compact electroencephalography (EEG) decoders enable embedded brain--computer interfaces (BCIs), but compression can alter adversarial robustness without proportional hardware savings. We present ORBITAL, a robustness-constrained framework combining hardware-realizable structured compression, recovery-aware operating-point selection, multi-attack evaluation, and native edge validation. Across nine subjects and three seeds on BCI Competition IV-2a, S25-TRADES achieves the highest observed mean true-union robustness among compressed EEGNet models (23.06% vs 21.27% for standard Full), with 26.27% fewer parameters and 25.10% fewer MACs. Matched controls identify TRADES as the main source of robustness recovery, improving S25 union accuracy by 6.74 percentage points over clean recovery (p=0.0039), while per-example analysis reveals boundary redistribution rather than uniform margin expansion. On Jetson Orin Nano, FP16 compression reduces board energy by up to 16.20% and latency by 8.00%. Despite 99.3% prediction agreement, native INT8 Q/DQ incurs about twice the FP16 latency and energy, demonstrating the need for joint robustness and hardware-aware optimization.
Daily item co-occurrence embeddings are paired with next-day behavior without future-behavior leakage. Multi-day click-through rate (CTR) training therefore mixes versions whose directions and norms may differ for the same item. We compare Word2Vec and an item-graph XSimGCL adaptation across initialization, inheritance, post-processing, training constraints, and downstream augmentation. The inventory contains 48 configurations, including 32 common-protocol CTR evaluations. Observed metrics are reported as relative percentage changes from the corresponding Word2Vec baseline (0%), with raw values and exact dataset sizes withheld. Three findings emerge. First, multiple upstream interventions improve compatibility and CTR. ID-init and orthogonal alignment increase cosine by 51.2% and 93.4%, and GAUC by 2.2% and 2.6%, relative to ordinary Word2Vec. Retrieval and geometry remain insufficient downstream surrogates. Second, warm starts perform strongly but accumulate scale. The day-matched norm increase over Word2Vec reaches 102.5% after eight days; daily global rescaling limits it to 8.2% and attains the highest observed GAUC increase, 4.0%. Third, CTR learning through a dimensionality-reducing encoder further adapts to residual drift. Against dimension-matched random-projection, SVD, and untrained-network controls, trained encoders attain the highest cross-day cosine for four of six upstream methods. The other two rank third and second in train-to-test cosine. These industrial results support complementary roles for upstream stabilization, inherited-scale management, and downstream task learning.
Monero is a privacy-focused cryptocurrency that deploys the Dandelion++ protocol and incorporates anonymity networks (such as Tor and I2P) to prevent malicious attackers from linking transactions with their source IPs. In this paper, we demonstrate that Monero's integration of the Tor network introduces a fundamental vulnerability: a Monero Tor node's originated transactions are exclusively forwarded to two outgoing Tor hidden service nodes (proxy nodes) prior to clearnet propagation, enabling an adversary to capture originated transactions by occupying the target node's outgoing connections. Based on this observation, we propose \textit{ProxyMark}, a three-stage deanonymization framework for the Monero Tor network, comprising node role identification, originated transaction identification, and node location deanonymization. Through experiments on the live Tor network, Monero mainnet, and testnet, we empirically demonstrate the effectiveness of \textit{ProxyMark} in successfully deanonymizing transactions originating from Monero nodes over Tor.
Srikanth Sivaramakrishnan, Behzad Danaei, Adwait Verulkar, Ryan Trinter
原文摘要
TrackJoint is a Modelica MultiBody component that turns a three-dimensional road surface into a joint with two degrees of freedom. The two joint coordinates are the distance travelled along a reference line and the sideways offset from that line. Together they give the position, orientation and angular velocity of a single frame_b connector. All road geometry reaches the joint as a small set of periodic piecewise-quintic splines, so one component can represent any circuit whose splines have been fitted beforehand. The road constraint is closed with d'Alembert's principle. The two generalized forces that belong to the joint coordinates are set to zero, so no Lagrange multipliers are needed. The joint is connected to a multibody race car in Dymola and the pair is exported as one FMI~3.0 FMU. The same FMU is then used for time-domain simulation and for CasADi/IPOPT minimum lap-time optimization, with no change to the model in between. This works because the joint keeps the arc-length coordinate as a real state and publishes its time derivative, which is the quantity that lets the problem be rewritten with distance We report results for minimum lap time simulations on the Singapore Formula~1 circuit.
Radio-frequency (RF) digital twins (DTs) enable site-specific wireless propagation prediction by explicitly reconstructing the physical environment and modeling electromagnetic interactions. Although physics-based DTs reduce reliance on exhaustive measurements, systematic bias relative to real radio data arises from environmental uncertainty, material inaccuracies, and geometric abstractions. Calibration using field measurements is therefore necessary but costly under limited sampling budgets. This paper studies active RF DT calibration, in which measurement locations are selected sequentially and in an online manner to estimate a calibration offset under strict measurement budget constraints. A physics-based RF DT pipeline is developed, integrating semantic geospatial data from OpenStreetMap, 3D modeling in Blender, and deterministic ray tracing in Sionna to capture dominant building and foliage effects. This RF DT is calibrated using path-loss measurements collected with software-defined radios. Multiple online calibration strategies that select radio measurement locations are compared, including random sampling, rank-structured sampling with balanced coverage, quasi-Monte Carlo ranked selection, progressive structure-aware allocation, and fully adaptive uncertainty-driven sampling with estimator stabilization. Results demonstrate that adaptive and structure-aware schemes significantly improve both mean and worst-case calibration accuracy, particularly in the low-budget regime for radio measurements.
Each new generation of frontier model pushes the edge of what is possible. Inside a harness -- a loop, a shell, a filesystem, the ability to write and run its own code -- a model can take on increasingly difficult work. A growing body of work treats the coding-agent harness as infrastructure for knowledge work rather than a coding tool, and a new wave of benchmarks tests it there. Most of them are generative: build a financial model, a memo, a spreadsheet from a brief, judged by a rubric or by another model. Few test the expensive part of knowledge work, which is investigative. In audit, tax, due diligence, regulatory reporting, claims or any dispute over numbers, the answer already exists -- hidden in incomplete and inconsistent evidence spread across systems -- and has to be recovered and defended to the cent. We introduce kw-bench, a benchmark built from real engagements specialists described to us. Its first task family is a five-year VAT reconstruction across three generations of SAP -- legacy ECC, a partial S/4HANA migration and a consultant's redesign -- delivered as raw system extracts: page headers, local number formats, partial link tables, no documentation of how anything connects. Grading is deterministic, with no LLM judge. We ran four harnesses -- Claude Code, Codex, OpenCode and micro-cc -- on the same model, DeepSeek v4.1 Flash, for two hours. The two independent, model-agnostic harnesses delivered (OpenCode 0.542, micro-cc 0.528); the two lab harnesses, away from their home model, delivered nothing (0.00 each). What defeated them was inertia: keeping many weak signals alive across a long session. On a frontier model the task collapses. micro-cc and Claude Code on Claude Opus 5.5 scored 0.9986 and 1.000, in 22 and 25 minutes. A weeks-long engagement now costs minutes and a few dollars; below the frontier, the harness decides how far a model gets.
Francesco Ortu, Joeun Yook, Punya Syon Pandey, Bernhard Schölkopf, Alberto Cazzaniga, Rada Mihalcea, Zhijing Jin
原文摘要
Large language models (LLMs) are increasingly used to disseminate information, yet their handling of contested historical events and politically charged narratives remains understudied. We introduce HistoricalMisinfo, a dataset of 500 contested events from 45 countries and regions, each paired with a factual and a documented revisionist reference narrative, instantiated across 11 prompt scenarios reflecting common real-world usage. Using an LLM-as-a-judge protocol, we evaluate multiple model architectures under neutral prompts and robustness prompts that explicitly request revisionist content. Under neutral prompts, models generally align with factual references, though alignment varies by scenario and model. Robustness prompting reveals a consistent and sharp increase in revisionism scores across all models, indicating limited resistance to user-driven revisionist framing. HistoricalMisinfo offers a practical benchmark for evaluating robustness to historical revisionism in AI systems. Code and dataset: https://github.com/francescortu/PreservingHistoricalTruth
Language models have recently solved open Erdős problems and achieved gold-medal level performance on International Mathematical Olympiad problems. But their solutions, written in natural language, cannot yet be verified by proof assistants. An erroneous reasoning step can invalidate every step that comes after, so it is advantageous to detect errors early. Several methods have been proposed to detect erroneous reasoning quickly based on the features of the forward pass of the language model that performs reasoning. In this work, we study whether the directed Forman curvature of that model's attention graph, a local quantity computed at an edge from the attention weights around it, carries information about a step's correctness. To do so, we evaluate four per-step verifiers on ProcessBench with and without the curvature appended to their input. Appending curvature values improves the overall mean performance of all four verifiers. ReProbe, the verifier that takes every token of a step as input, shows the largest improvement. Adding the curvature increases the time of the language model's forward pass by 18% on average. Our code is available at https://github.com/kanghyeonz/attention-graph-curvature.
Big goals are hard to achieve all at once; breaking them into small steps is wiser. We present Trust Region Policy Distillation (TOP-D), which transforms the notoriously unstable, high-variance On-Policy Distillation (OPD) into a stable training paradigm by dynamically constructing a proximal teacher. Theoretically, we establish a rigorous framework demonstrating that TOP-D inherently controls gradient variance. By providing a formal global convergence analysis alongside a monotonic improvement bound, we mathematically formalize the reliability and stability of the overall training dynamics. Empirically, TOP-D dramatically enhances training stability, sample efficiency, and final performance on mathematical reasoning tasks. More importantly, TOP-D introduces zero additional computational overhead, positioning itself as a promising alternative to the well-established OPD paradigm.
One of the most pressing challenges in artificial intelligence is to make models more transparent to their users. Recently, explainable artificial intelligence has come up with numerous methods to tackle this challenge. A promising avenue is to use concept-based explanations, that is, high-level concepts instead of plain feature importance scores. Among this class of methods, Concept Activation Vectors (CAVs, Kim et al., 2018) stand out as one of the main protagonists. One interesting aspect of CAVs is that their computation requires sampling random examples from the train set. Therefore, the actual vectors obtained may vary depending on the randomness of this sampling. In this paper, we propose a fine-grained theoretical analysis of CAV construction in order to quantify their variability. Our results, confirmed by experiments on several real-life datasets of four different modalities, point to an universal result: the variance of CAVs declines roughly as $1/N$, where $N$ is the number of random examples. Based on this, we give practical recommendations for a resource-efficient application of the method.
Automatic differentiation (the automatic computation of numerical function derivatives) has many applications, especially in scientific computing and machine learning. Several approaches to automate the calculation of derivatives have been presented. Common challenges facing those approaches include (1) having to define derivative rules for different classes of functions, (2) having to manipulate program symbols to understand the semantics of the function being differentiated, and construct the derivative function, and (3), more importantly, ensuring that the calculated derivative is correct. In this paper, we present an approach for automatic \emph{certified} differentiation using the Lean proof assistant. By certified, we mean a correctness certificate is computed together with the derivative function and a domain on which the derivative is guaranteed to be valid. Our approach leverages existing derivative theorems included in the Mathlib Lean library instead of having to define our own rules. We also utilize Lean metaprogramming to manipulate the symbolic structure of the input function, and to construct the derivative function. The correctness certificate is computed step-wise as we apply Mathlib theorems. To make the search efficient, we index the derivative theorems by the function they differentiate. We evaluate our approach on 500 randomly generated functions: with the index, all of them are differentiated, with a median time of 0.18\,s per function, 8.6 times faster than without the index.
Self-driving laboratories couple a learned design surrogate to an instrument, and on every new target domain it must answer four questions: when it can be trusted, how many target measurements it needs, which to make first, and when to stop adapting, switch models or escalate. We state a framework making each measurable (a competence ratio, a budget $k^*(\varepsilon)$, a time-to-target and a priced decision) and instantiate it without wet experiments on six public droplet-microfluidic and liposome tables as offline laboratories: 107 leave-one-group tiers whose rows are hidden labels. A sealed, pre-registered programme (two sealed negative validations and a pre-registered test on three more tables) and seven frozen closed-loop protocols, each run twice, yield 23 findings. Shift cost is stratified by shift type; no label-free statistic certifies individual candidates; random-split model-family rankings predicted neither the sealed nor the loop rankings. Warm-start residual boosting returns a surrogate to $1.25\times$ its development reference within 20 target rows in 94% of chemistry tiers (27% under a strict noise-floor reading), 90% of laboratory and 40% of device tiers, and a budget planned from the shift type on other tables is not measurably worse on a new one. In the loop, one gain survives drift and shuffled-label controls on all tiers, carried by the sparsely sampled ones: proposing at once and warm-starting on every measurement reaches a 10% target in 2.80 measurements against 3.31 for the reference, and in 5.72 against 11.35 for random proposals on 301 hard units at a 3% tolerance. No label-free acquisition rule survives the shuffled-label control, no interval-aware proposal or other model family does better, and no threshold rule on the tested diagnostics beats always continuing once escalation is priced. No wet-lab validation is claimed.
Alexander Nemecek, Osama Zafar, Yuqiao Xu, Wenbiao Li, Erman Ayday
原文摘要
Watermarking is becoming the default mechanism for AI content authentication, with governance policies and frameworks referencing it as infrastructure for content provenance. Yet across text, image, and audio modalities, watermark signal strength, detectability, and robustness depend on statistical properties of the content itself, properties that vary systematically across languages, cultural visual traditions, and demographic groups. We examine how this content dependence creates modality-specific pathways to bias. Reviewing the major watermarking benchmarks across modalities, we find that, with one exception, none report performance across languages, cultural content types, or population groups. To address this, we propose three concrete evaluation dimensions for pluralistic watermark benchmarking: cross-lingual detection parity, culturally diverse content coverage, and demographic disaggregation of detection metrics. We argue that watermarking is part of the pluralistic alignment pipeline and should be held to the same evaluation standards. We connect this to governance frameworks currently mandating watermarking deployment without requiring fairness evaluation. Our position is that evaluation must precede deployment, and that the same bias auditing requirements applied to AI models should extend to the verification layer.
Zeyuan Feng, Ali Fuat Sahin, Santiago Thorup, Somil Bansal
原文摘要
Hamilton-Jacobi (HJ) reachability provides a principled framework for synthesizing safety certificates and robust controllers for safety-critical robotic systems. However, applying reachability analysis to high-dimensional nonlinear systems remains challenging: classical grid-based solvers suffer from the curse of dimensionality, continuous-time neural solvers require accurate spatial value gradients, and reinforcement-learning-based approaches often suffer from weak boundary anchoring and non-stationary adversarial policy optimization. We propose a discrete-time neural reachability framework for control-disturbance-affine systems that learns backward reachable tubes (BRTs) and backward reach-avoid tubes (BRATs) through Bellman-Isaacs value propagation. Our key idea is to combine equation-driven self-supervision with structured policy learning: rather than computing explicit PDE-gradients, we exploit the bang-bang structure of optimal safety interventions to construct approximate teacher actions from gradient-free value probes, converting adversarial actor learning into supervised policy learning. To stabilize long-horizon value propagation, we leverage the learned actor to train the value function backward from the terminal boundary using a windowed temporal curriculum, where each window is used as the boundary condition for the next window. Across benchmark problems up to 80 dimensions, our method learns accurate reachability value functions while improving stability over existing learning-based solvers. We further demonstrate observation-space scalability on F1-tenth racing with over 16,000-dimensional egocentric inputs. The learned safety filter generalizes zero-shot to unseen tracks and transfers to a physical RC car, achieving real-time robust collision avoidance.
Saul D. Freedman, Melissa Lee, Dragan Marušič, Tomaž Pisanski
原文摘要
Given a finite graph $G$, we characterize the automorphisms and automorphism groups of $G$ that can be realized as geometric symmetries and symmetry groups of straight line drawings of $G$ in the plane. In doing so, we correct a previous version of this characterization from the literature. As an application, we show that the graphs recently constructed by the first two authors to disprove the Polycirculant Conjecture, which have orders $16{,}464^k$ for each positive integer $k$, are the first known vertex-transitive graphs with no non-trivial realizable geometric symmetries. We also construct infinite families of graphs that exhibit all possible kinds of geometric symmetry.
Modeling multi-scale dynamics in urban road networks is essential for intelligent transportation systems but is often hindered by the computational cost of long-range historical context and the challenge of fusing heterogeneous modalities. Existing spatio-temporal frameworks face a trade-off between sequence length and memory overhead, while frequently neglecting the transient effects of exogenous factors such as road incidents and weather. To address these limitations, we propose SpecLST, a frequency-enhanced long-short-term spatio-temporal multimodal framework. SpecLST leverages a fast Fourier transform-based frequency-domain representation module to efficiently encode multi-week historical dependencies, bypassing the memory bottlenecks of standard attention mechanisms over raw long sequences. Heterogeneous context is integrated through attention-based fusion, followed by a sparse Mixture-of-Experts backbone with graph augmentation for spatio-temporal representation learning. Additionally, a frequency-enhanced residual head isolates high-frequency transients, improving sensitivity to abrupt traffic anomalies. At the 60-minute horizon, SpecLST achieves mean relative error reductions of 14.10% in MAE, 10.35% in RMSE, and 15.78% in MAPE across eight neural forecasting baselines and four datasets, with the dataset- and metric-specific average reductions ranging from 5.63% to 22.21%.
Reza Yousefi Maragheh, Jayesh Uddhav Kudase, Aysenur Inan, Ramin Giahi, Kai Zhao, Parth Parikh, Jianpeng Xu, Jason Cho 等 10 位作者
原文摘要
Evaluating natural language explanations in recommender systems is essential for fostering user trust, transparency, and engagement. However, existing evaluation approaches like human evaluations, while accurate, are resource-intensive and impractical at the scale required by modern recommendation platforms. Also, automated methods using single-agent LLMs suffer from prompt sensitivity and inconsistent outputs. To address these challenges, we propose MAREval, a structured multi-agent framework for evaluating recommendation explanations using large language models. MAREval orchestrates (i) a planner agent that uses a novel Chain of Debate (CoD) prompting strategy to coordinate agent roles and enforce logically consistent evaluation plans; (ii) a moderator agent that regulates discussions by mitigating prompt drift; and (iii) an arbitrator agent that aggregates outputs from multiple evaluation rounds and (iv) a Monte Carlo sampling method, improving robustness and alignment with human judgment. We conduct comprehensive evaluations on both public (TopicalChat) and proprietary recommendation datasets, demonstrating that MAREval outperforms state-of-the-art baselines. Comprehensive experiments on a public benchmark and a proprietary e-commerce dataset show that MAREval improves alignment with human judgments over strong singleand multi-agent baselines. Stability analyses indicate substantially lower variability across repeated trials. In a large human-annotation gate, MAREval meets production quality thresholds where prior evaluators fall short, and online A/B testing demonstrates statistically significant improvements in engagement and revenue metrics. These results establish MAREval as a scalable and reliable solution for human-aligned evaluation of recommendation explanations in real-world systems