arXiv:2606.08328v1 Announce Type: new Abstract: Equitable allocation of indivisible goods to agents in online settings is an algorithmic primitive with applications for load balancing, network routing, online marketplaces, and multi-agent systems. We consider a general setting in which allocations are constrained to be bases of discrete polymatroids that arrive online. Our work demonstrates that a simple, myopic algorithm called Brick-Laying, which greedily minimizes the sum of squared loads on agents, achieves a universal and objective-free notion of optimality called majorization minimax-optimality [BDK26] for this setting. As a consequence, Brick-Laying simultaneously guarantees minimax optimal competitive ratios and regret for all Schur-concave and Schur-convex objectives, and for any number of agents and resources (despite being agnostic to problem scale). Departing from popular primal-dual analysis, we employ majorization to compare allocations. We leverage the conjugates of integer partitions -- which act as a discrete dual to majorization -- to characterize worst-case instances for the Brick-Laying algorithm. Our approach reveals a novel structural connection between the geometry of partitions and online equitable allocation.
Science Journals
arXiv:2606.08256v1 Announce Type: new Abstract: Verifiability, attribution, and reproducibility are foundational requirements of scientific knowledge, yet current publishing infrastructure does not enforce them at scale. We introduce Traxia, an agent-native scientific publishing framework in which AI research agents publish verifiable papers, build reputational identities, peer-review one another, and collaborate with humans in a shared provenance model. Traxia treats agents as first-class epistemic participants: every paper carries a reasoning trace, every claim a confidence interval, every agent a cryptographically signed identity, and every collaboration an immutable contribution log. We formalise five components: Agent Identity and Registry, Verifiable Publishing Layer, four-tier Peer Review Protocol, Reputation and Staking Engine, and a Knowledge Graph with contradiction detection. The framework targets reproducibility failure, provenance opacity, and exclusion of Global South research capacity. This paper presents architectural foundations and formal specifications only; it does not report empirical results. Evaluation and deeper component studies will follow in subsequent papers. A prototype partially implements core formalisms; the full system remains under active development.
arXiv:2605.06582v2 Announce Type: replace Abstract: Many operations on sensory data -- comparison, memory, retrieval, and reasoning -- are naturally expressed over discrete symbolic structures. In language this interface is given by tokens; in audio, it must be learned. Existing audio tokenizers rely on quantization, clustering, or codec reconstruction, assigning tokens locally, so sequence consistency, compactness, length control, termination, and edit similarity are rarely optimized directly. We introduce PairAlign, a framework for compact audio tokenization through sequence-level self-alignment. PairAlign treats tokenization as conditional sequence generation: an encoder maps speech to a continuous condition, and an autoregressive decoder generates tokens from BOS, learning token identity, order, length, and EOS placement. Given two content-preserving views, each view's sequence is trained to be likely under the other's representation, while unrelated examples provide competing sequences. This gives a scalable surrogate for edit-distance preservation while discouraging many-to-one collapse. PairAlign starts from VQ-style tokenization and refines it with EMA-teacher targets, cross-paired teacher forcing, prefix corruption, likelihood contrast, and length control. On 3-second speech, PairAlign learns compact, non-degenerate sequences with broad vocabulary usage and strong cross-view consistency. On retrieval tests, it preserves edit-distance search while reducing archive token count by 55%. A continuous-sweep probe shows lower local overlap than a dense geometric tokenizer, but stronger length control and bounded edit trajectories under 100 ms shifts. PairAlign is a sequence-symbolic predictive learner: like JEPA-style objectives, it predicts an abstract target from another view as a learned variable-length symbolic sequence, not a continuous latent.
arXiv:2605.08384v3 Announce Type: replace Abstract: In this work, we introduce GELATO (Geometry-preserving Embeddings via Locked Aligned TOwers), a novel approach to multimodal embedding models. We build on the VLM-style architecture, in which non-text encoders are adapted to produce input for a language model, which in turn generates embeddings for all varieties of input. We present the result: the jina-embeddings-v5-omni suite, a pair of models that encode text, image, audio, and video input into a single semantic embedding space. GELATO extends the two Jina Embeddings v5 Text models to support additional modality by adding encoders for images and audio. The backbone text embedding models and the added non-text modality encoders remain frozen. We only trained the connecting components, representing 0.35% of the total weights of the joint model. Training is therefore much more efficient than full-parameter retraining. Additionally, the language model remains effectively unaltered, producing exactly the same embeddings for text inputs as the Jina Embeddings v5 Text models. Our evaluations show that GELATO produces results that are competitive with the state-of-the-art, yielding nearly equal performance to larger multimodal embedding models.
arXiv:2605.09823v3 Announce Type: replace Abstract: Personal AI assistants are beginning to act as delegates with access to calendars, inboxes, and user preferences. Calendar scheduling makes the trust problem concrete: an assistant must coordinate with other assistants while deciding what to reveal about the person it represents. We introduce CalBench, a controlled benchmark for multi-agent calendar scheduling under private information. In each task, $N$ agents manage separate private calendars and schedule a stream of $M$ incoming meetings while minimizing disruption costs. Because no agent can inspect another agent's calendar, success requires language-mediated coordination rather than centralized planning. CalBench generates solvable scenarios with CP-SAT oracle solutions and decentralized non-LLM reference protocols, enabling evaluation of task success, excess cost, communication efficiency, burden fairness, and privacy leakage under matched information constraints. Across seven model families, we find that completion alone misses important failures: agents leave avoidable cost on the table, communication volume does not predict lower regret, and privacy-preserving silence can deprive teammates of cost information needed for fair burden allocation. CalBench provides a reproducible testbed for studying whether autonomous assistants can coordinate on behalf of users before deployment at scale.
arXiv:2605.11314v3 Announce Type: replace Abstract: Cerebral Palsy (CP) is a neurological disorder of movement and the most common cause of lifelong physical disability in childhood. Approximately 75% of children with CP are ambulatory, and accurate gait assessment is central to preserving walking function, which deteriorates by mid-adulthood in a quarter to half of adults with CP. The Rodda and Graham classification system quantifies sagittal-plane gait deviations using ankle and knee z-scores derived from 3D Instrumented Gait Analysis (3D-IGA), but 3D-IGA is expensive and limited to specialized centers, while observational assessment shows only moderate inter-rater agreement. We developed a markerless gait analysis pipeline that quantifies Rodda and Graham knee and ankle z-scores directly from single-view clinical gait videos. Across 1,058 bilateral limb samples from 529 trials of 152 children (88 male, 63 female; age 12.1 $\pm$ 4.0 years; 60 distinct primary diagnoses, cerebral palsy the most common at $n=54$), the sagittal-view model achieved $R^2 = 0.80 \pm 0.02$ and CCC $= 0.89 \pm 0.02$ for knee z-scores and $R^2 = 0.57 \pm 0.02$ and CCC $= 0.72 \pm 0.02$ for ankle z-scores against 3D-IGA. Binary screening for excess knee flexion achieves AUROC $= 0.88$, correctly identifying 83% of affected children, and applying Rodda and Graham rules yields $43 \pm 1$% 7-class accuracy with macro-AUROC $= 0.78 \pm 0.01$, ankle prediction error remaining the primary bottleneck. Beyond cross-sectional screening, continuous z-scores support longitudinal trajectory tracking across visits, providing a quantitative substrate for monitoring disease progression and treatment response unavailable from observational scales. These results demonstrate the feasibility of video-based z-score estimation, excess-flexion screening, and longitudinal trajectory tracking as a path toward scalable, objective gait assessment in low-resource clinical settings.
arXiv:2606.08777v1 Announce Type: new Abstract: Visual Language Models (VLMs) are known to produce hallucinated predictions that are not grounded in visual evidence, yet existing approaches lack a principled understanding of how robust such predictions are under counterfactual perturbations. In this work, we study the sample complexity of counterfactual robustness for hallucinated outputs in VLMs. We define a causal influence metric based on log-probability differences between factual, counterfactual, and activation-patched runs, and use it to characterize the stability of hallucinated predictions. By leveraging circuit discovery techniques (CD-T), we identify model components responsible for these predictions and track their activation differences across counterfactual samples. We then derive empirical bounds on the minimum number of counterfactual samples m required to reliably detect instability in hallucinated outputs, using concentration inequalities and variance estimates of the causal influence distribution.
arXiv:2404.01948v3 Announce Type: replace Abstract: Dynamic visual sensors (DVS) are characterized by a large amount of background activity (BA) noise, which it is mixed with the original (cleaned) sensor signal. The dynamic nature of the signal and the absence in practical application of the ground truth, it clearly makes difficult to distinguish between noise and the cleaned sensor signals using standard image processing techniques. In this letter, a new technique is presented to characterise BA noise derived from the Detrended Fluctuation Analysis (DFA). The proposed technique can be used to address an existing DVS issues, which is how to quantitatively characterised noise and signal without ground truth, and how to derive an optimal denoising filter parameters. The solution of the latter problem is demonstrated for the popular real moving-car dataset.
arXiv:2409.18867v2 Announce Type: replace Abstract: We propose a robust and efficient data-driven predictive control (eDDPC) scheme which is more sample efficient (requires less offline data) compared to existing schemes, and is also computationally efficient. This scheme employs a recently proposed data-based representation of linear time-invariant (LTI) systems as a predictor. Such a representation serves as an alternative to Hankel-based predictors obtained from, e.g., the so-called fundamental lemma, and can be derived by exploiting the kernel structure of shallow Hankel matrices of data. This allows for application of our proposed scheme using very short (and potentially irregularly measured) noisy input-output data, the amount of which is independent of the prediction horizon. To account for measurement noise, we provide a novel result that quantifies the uncertainty between the true (unknown) restricted behavior of the system and the estimated one from noisy data. Furthermore, we show that the robust eDDPC scheme is recursively feasible and that the resulting closed-loop system is practically exponentially stable. Finally, we compare the performance of this scheme to existing ones on a case study of a four tank system.
arXiv:2606.08403v1 Announce Type: new Abstract: Text-centered prompt-injection defenses assume that the malicious signal is visible in one of the inspected text views. We study a reproducible LLM01-style indirect prompt/content-injection failure mode where that assumption breaks: a payload caught in plain English slips past the same detector when it is transported as structured float parameters and reconstructed only as fragmented telemetry. Across 14,400 attacked real-model trials on three commercial LLM APIs from different providers, the IFS-derived float-array carrier preserves 94.3% leakage ASR under the strongest dual-layer text-classifier defense evaluated in the main matrix: a Prompt Guard 2 + TF-IDF ensemble; the same carrier-level pattern also replicates with a fine-tuned roberta-base detector. We emphasize leakage ASR because downstream systems may act on quoted or reproduced markers even when the model refuses, but Strong ASR is the stricter metric for structurally compliant attack success. A 2 x 2 ablation shows that data-layer storage and reconstruction-layer fragmentation defeat different text views and that both are needed to evade both. A simple xxd detector and semantic validation block the current T3 instance, so the contribution is not an undetectable exploit but a measured failure boundary for text-only inspection in structured-input pipelines that expose reconstructed auxiliary channels to an LLM.
arXiv:2606.08412v1 Announce Type: new Abstract: Given a finite set of integers $A$, a \emph{unary translocation} produces a new set $A' = A \cup \{u,v\}$, where $u$ and $v$ are nonnegative integers satisfying $x+y=u+v$ for some $x,y\in A$. For an input set $A$ and a target set $B$, the \emph{unary translocation distance} is the minimum number of unary translocations required to obtain a superset containing $B$. In this paper, we study this problem from both theoretical and computational perspectives. We prove that computing the unary translocation distance is strongly NP-hard, thereby answering an open question raised by \citet{ConstantinMiclausPopa2026UnaryTranslocation}. On the positive side, we give an exact pseudo-polynomial algorithm for every fixed constant value of $|B|$, extending our previous results for $|B|\leq 2$. For arbitrary target sets, we present a $2$-approximation algorithm, an additive $(|B|-1)$-approximation algorithm, and show that the additive algorithm also yields a $3$-approximation. We also propose parameterized algorithms, including algorithms parameterized by the maximum value in the input set together with the optimum distance, and by the maximum value in the target set together with $|B|$. In addition, we propose an integer linear programming formulation that gives an exact mathematical model for the problem, analyze its size, and show that the LP relaxation has integrality gap at least $\frac{4}{3}$. Finally, we report computational experiments comparing the $2$-approximation algorithm, beam search, and simulated annealing. The results show that the approximation algorithm is highly effective in practice and often outperforms the heuristic baselines.
arXiv:2606.08646v1 Announce Type: new Abstract: A multivariate polynomial on $n$ variables $x_1,\ldots,x_n$ of total degree $n$ over $\mathbf{Z}_2$ containing the multilinear monomial $\prod_{i=1}^n x_i$ is by the combinatorial nullstellensatz [Alon, Comb. Probab. Comput., 1999] known to always have a nonroot. We show that there cannot be a randomised polynomial time algorithm that given an arithmetic circuit of polynomial size formally computing such a polynomial, locates a nonroot with constant nonzero probability unless RP=NP. The result holds even when the individual degree of every variable in the input polynomial is at most two.
arXiv:2606.09273v1 Announce Type: new Abstract: 3D semantic scene generation is crucial for autonomous driving applications, yet most methods rely on complex 3D-specific architectures such as triplane encoders and adapted diffusion networks, limiting both their simplicity and their editing capabilities. We propose EditSSC, an editing-ready method for 3D semantic scene generation using 2D Bird's Eye View (BEV) representations and off-the-shelf latent diffusion network. Our approach reshapes 3D semantic occupancy grids into multi-channel BEV images and leverages the quantized autoencoder and UNet from Stable Diffusion with minimal modifications. We perform diffusion on the latents after quantization, which enables training-free editing capabilities. By exploiting class-to-code correspondences in the codebook, our method supports sketch-guided generation, inpainting, and outpainting without any retraining. On SemanticKITTI, EditSSC outperforms existing 3D-specific baselines on unconditional generation, demonstrating that well-established 2D architectures can be effectively repurposed for 3D scene generation and editing.
arXiv:2606.08444v1 Announce Type: new Abstract: Large Language Models (LLMs) have become powerful tools for code generation, yet they remain prone to hallucinations-producing plausible but incorrect or fabricated outputs. Among these, package hallucination, where an LLM suggests non-existent dependencies, poses an emerging security risk to the software supply chain. While previous studies focus on popular languages like Python or JavaScript, in this work we present the first large-scale empirical study on crate hallucination in LLM-generated Rust code. We construct a multi-source dataset combining coding tasks from Stack Overflow, GitHub, and LLM-generated tasks, and evaluate both commercial and open-source models under various decoding settings. Our analysis reveals that, unlike prior findings in Python and JavaScript, hallucination behavior in Rust follows a distinct pattern: different models exhibit surprisingly consistent hallucination rates, and these rates show minimal sensitivity to model parameters. Furthermore, we investigate prompt engineering strategies to mitigate hallucinations without sacrificing code quality. This study provides new insights into the reliability and security implications of LLM-assisted Rust development, offering guidance for future research and safer model deployment in software engineering workflows.
arXiv:2606.08450v1 Announce Type: new Abstract: Financial portfolio trading is naturally formulated as a reinforcement learning problem, where an agent sequentially rebalances assets under changing market conditions to balance return, risk, and transaction costs. Yet in non-stationary markets, raw OHLCV states and short-horizon return rewards often provide an under-specified learning interface, motivating large language models as a way to inject financial knowledge into state and reward design while constraining open-ended generation. To this end, we propose GIFT, an LLM-guided framework for state-reward interface design in PPO-based financial reinforcement learning. Rather than using the LLM to make trading decisions, GIFT uses Factor-guided State Enhancement to generate state features from financial-factor primitives, Risk-rule-guided Reward Shaping to generate auxiliary rewards from portfolio-risk rules, and Diagnostic-guided Refinement to revise candidate interfaces using PPO rollout diagnostics. After refinement, GIFT fixes the selected state-reward interface before evaluation, with no further LLM queries or interface updates at test time. Comprehensive rolling-window experiments across diverse market regimes and portfolio scenarios demonstrate that GIFT improves learning-signal quality and out-of-sample risk-adjusted portfolio performance over baselines. Code and data are available at: https://github.com/KAG778/GIFT .
arXiv:2501.06659v2 Announce Type: replace Abstract: Many templatized documents are programmatically generated from structured data following a visual template. Such documents include invoices, tax documents, financial reports, and purchase orders. Effective data extraction from these documents is crucial to support downstream analytical tasks. Current data extraction tools often struggle with complex document layouts, incur high latency and/or cost on large datasets, and require significant human effort. The key insight of our tool, TWIX, is to infer the underlying template used to create such documents, and then extract the data, rather than extracting directly from documents. To do so, TWIX first infers the underlying fields, such as columns of tabular portions or keys in co-located key-value pairs, by leveraging their consistent location patterns (e.g., two fields in the same template repeatedly co-occur within a fixed distance apart across multiple records). TWIX then assembles these fields into a template by enforcing visual constraints, such as vertically aligning table rows with their column headers for tabular regions, and horizontally aligning keys with their values for key-value pairs. TWIX then uses this inferred template to accurately and efficiently extract data from templatized documents at a low cost. On one benchmark with 34 diverse real-world datasets, TWIX outperforms state-of-the-art structured data extraction tools (Evaporate, Textract, and Azure Document Intelligence), and vision-based LLMs like GPT-4-Vision, by over 25% in precision and recall. Another benchmark with 30 large datasets demonstrates TWIX's scalability: it is 520X faster and 3,786X cheaper than the most competitive compared tool, for extracting data from large document collections with over 2000 pages.
arXiv:2502.06819v2 Announce Type: replace Abstract: This paper presents a framework for generating 3D indoor scenes from text prompts. Existing methods often formulate scene synthesis as an object layout prediction problem conditioned on a single input modality, such as a text description, room shape, or scene graph. This design can lead to object collisions and limited functional plausibility, reducing its practical applicability. To address these limitations, we introduce a multi-stage pipeline that better reflects practical scene creation scenarios. Given a text prompt describing partial scene content, our method first uses graph diffusion to produce a contextually coherent scene graph and then predicts a realistic object layout. In addition, we incorporate lightweight human-object interaction priors to encourage human-centric and functional arrangements, with explicit spatial constraints to reduce interpenetration. Our approach generates coherent 3D scenes with viable layouts that better support human interaction. Experiments on the 3D-FRONT dataset demonstrate that our method achieves competitive or state-of-the-art performance compared with existing approaches, while improving the physical plausibility of generated scenes.
arXiv:2605.15416v2 Announce Type: replace Abstract: Jung et al. (2025) introduce a hypothesis testing framework for guaranteeing agreement between large language models (LLMs) and human judgments, relying on the assumption that the model's estimated confidence is monotonic with respect to human-disagreement risk. In practice, however, this assumption may be violated, and the generalization behavior of the confidence estimator is not explicitly analyzed. We mitigate these issues by learning a dedicated confidence estimator instead of relying on heuristic confidence signals. Our approach leverages simulated annotator diversity and a margin-based ranking formulation to explicitly model how confidently an LLM distinguishes between human-agreement and human-disagreement cases. We further derive generalization guarantees for this estimator, revealing a margin-dependent trade-off that informs the design of an adaptive estimator training procedure. When integrated into fixed-sequence testing, the learned confidence estimator yields improved ranking accuracy and empirically strengthens the monotonic relationship between confidence and disagreement risk, leading to higher success rates in satisfying target agreement levels across multiple datasets and judge models.
arXiv:2606.08158v1 Announce Type: new Abstract: Large language models (LLMs) can generate factually inconsistent claims, motivating accurate and scalable hallucination detectors. Prior work largely enlarges training sets via synthesis or new annotations, introducing increasing cost and potential bias while underusing the consistency implied by semantically equivalent paraphrases. We propose Consistency-Constrained Hallucination Detector (CCHD), which formulates training as a constrained optimization problem. The standard cross-entropy on original document-claim pairs is complemented by (i) paraphrase-consistency constraints bounding divergence across paraphrased views, and (ii) label-preservation constraints tying paraphrases to ground truth. We solve the problem by gradient descent-ascent over model parameters and per-view Lagrange multipliers, adding only a few scalar dual variables and no inference-time overhead. With DeBERTa and Flan-T5 backbones, CCHD consistently outperforms strong baselines (FactCG, MiniCheck, and AlignScore) on standard factuality benchmarks, demonstrating its superiority on hallucination detection.
arXiv:2606.08467v1 Announce Type: new Abstract: While confidence calibration is essential for trustworthy decision-making in safety-critical applications, the robustness of calibrated GNNs to adversarial structural perturbations remains largely unexplored. However, studying calibration attacks on graphs presents unique technical challenges: (1) the discrete nature of graph structures complicates gradient-based optimization, (2) existing underconfidence objectives fail to drive predictions toward uniform distributions, and (3) GNNs are highly sensitive to edge perturbations, often causing unintended label changes that violate attack constraints. To address these challenges, we propose a \textbf{Unified Graph Calibration Attack (UGCA)} framework designed for \textbf{worst-case (white-box) analysis} of GNN calibration robustness. UGCA introduces a KL-divergence loss to encourage uniform predictive distributions, a reranking mechanism to reduce label flipping, a hybrid loss to recover labels when violations occur, and beam search to explore a broader adversarial search space. We further provide theoretical insights linking model generalization, dataset complexity, and calibration vulnerability, showing that models with higher accuracy or trained on datasets with more classes are more susceptible under this threat model. Extensive experiments demonstrate that UGCA substantially increases Expected Calibration Error while preserving classification accuracy. Our code is publicly available at https://github.com/CaptainCuong/Graph-Calibration-Attack.git.
arXiv:2606.08476v1 Announce Type: new Abstract: Context parallelism (CP) is essential for training large-scale, long-context language models, as it partitions sequences to reduce memory overhead. However, existing CP methods suffer from workload imbalance, inefficient kernels, and redundant communication due to static sequence sharding and key-value (KV) tensor communication. We present FlashCP, a load-balanced and communication-efficient framework for CP training. FlashCP introduces a sharding-aware communication mechanism to eliminate redundant KV communication and proposes a novel Whole-Doc sharding strategy that maximizes communication savings while maintaining balanced workloads. To efficiently combine Whole-Doc and Per-Doc sharding, FlashCP further designs a heuristic algorithm to search for near-optimal sharding plans. Extensive experiments show that FlashCP achieves up to 1.63x speedup over state-of-the-art CP frameworks across diverse datasets.
arXiv:2606.09689v1 Announce Type: new Abstract: We develop a numerical algorithm for the efficient solution or approximation of solutions to the Helmholtz equation on a structured grid in two dimensions. We make use of the Operator Fourier Transform (OFT) and a low-rank cross approximation scheme (Cross-DEIM) to decompose the problem into an integral over a pseudo-time of solutions to the Schr\"odinger equation. The OFT is a framework for solving operator equations like fractional Laplacian equations or the Helmholtz equation, when the latter is written as a product of two paraxial operators. The main computational cost in the OFT is the solution to the Schr\"odinger equation, especially when the dimension or mesh resolution is high. In this work, we alleviate this cost by utilizing a low-rank method. Such methods aim to beat the curse of dimensionality when low-rank structures are present in the solution. We show that the combination of these two approaches can have large cost reductions for certain classes of problems.
arXiv:2606.08198v1 Announce Type: new Abstract: Despite their growing popularity, in-ear Earable / Hearable devices (i.e., ear-mounted wearables) face interaction challenges due to limited input space and compact form factors. To enhance interaction capabilities, researchers are exploring off-device hand-based input spaces above the neck using midair and onskin gestures. However, existing literature primarily focuses on axial swipes (i.e., horizontal and vertical), leaving nonaxial swipes (i.e., unidirectional swipes with varied orientations) and angular swipes (e.g., L, U, or V) largely underexplored despite their potential interaction advantages. To address this gap, we conducted a within-subject gesture motion analysis study with 24 participants, analyzing 5,568 swipes of varying shape, orientation, and complexity. Our results revealed preferred starting and ending regions for different unidirectional and angular swipe shapes, as well as intuitive swipe shapes within the off-device, above-neck manual interaction space. We further examine off-device swipe characteristics, discuss the feasibility of recognizing these earable gestures with current sensing technologies, and highlight their potential application in various scenarios. These findings broaden the understanding of off-device earable gestures and provide design insights for integrating suitable nonaxial and angular swipes alongside traditional axial gestures to enhance interaction with in-ear earable devices.
arXiv:2606.08800v1 Announce Type: new Abstract: In high-stakes settings such as brand compliance, clinical care, and content moderation, machine learning cannot be deployed as opaque oracles: practitioners inspect the features driving model decisions, and models must leverage the expert documentation governing these domains. In practice, the data arrives as unstructured content, and features extracted from it must be interpretable, discriminative, and aligned with what experts consider important. Existing methods fall short: they target tabular inputs, lack demonstrated expert alignment, and cannot operationalize qualitative criteria such as 'maintain professional tone' into precise features. We present FEST (Feature Engineering with Self-evolving Trees), combining dual-stream feature generation (semantic and deterministic), semantic deduplication, and tree-guided iterative evolution to discover auditable features from raw text and images. FEST leads in 17 of 20 classifier-task combinations across brand classification, content authenticity detection, and stress detection, with a mean gain of 4.2 pp over the strongest baseline across five classifiers. An LLM-as-judge evaluation shows FEST achieves 60-80% coverage of expert-designed brand features at strict semantic-alignment thresholds, corroborated by a human expert study rating features highly on relevance, clarity, and actionability. When seeded with expert guidelines, FEST refines qualitative criteria into operational features, improving accuracy by 6-12 pp on average across brands. To enable systematic evaluation of expert alignment in automated feature engineering, we release BrandGuide, the first dataset pairing expert-designed features with 1M+ assets across 2,683 brands. By grounding feature engineering in expert knowledge, FEST opens a practical pathway for interpretable ML in domains demanding human oversight.
arXiv:2606.08806v1 Announce Type: new Abstract: Artificial Intelligence (AI) and Large Language Models (LLMs) are increasingly used in autonomous software testing; however, AI-generated test artifacts often suffer from hallucinations, compliance violations, security risks, and limited explainability. To enhance the reliability, transparency, and trustworthiness of AI-generated testing artifacts, this research introduces the concept of Governance-Aware Autonomous Testing Framework (GATF). The framework extends the autonomous testing lifecycle with governance validation, explainability analysis, probabilistic risk assessment, compliance monitoring, as well as audit governance. Experiments were performed with Defects4J and PROMISE software engineering datasets. The proposed framework successfully reduced the governance-related risks by 89.6% and demonstrated 94.3% accuracy in governance, 96.5% artifact reliability, 94.2% compliance accuracy, and 90.8% explainability performance. The results show that autonomous testing systems that are governance-aware can significantly enhance the reliability, transparency, and operational security of autonomous testing systems in comparison to conventional AI-based testing systems. The proposed architecture is scalable and reliable and provides a safe environment for software testing.