arXiv:2502.16174v4 Announce Type: replace
Abstract: Although modern LLMs are aligned with human values during post-training, robust moderation remains essential to prevent harmful outputs at deployment time. Existing approaches suffer from performance-efficiency trade-offs and are difficult to customize to user-specific requirements. Motivated by this gap, we introduce Multi-Layer Prototype Moderator (MLPM), a lightweight and highly customizable input moderation tool. We propose leveraging prototypes of intermediate representations across multiple layers to improve moderation quality while maintaining high efficiency. By design, our method adds negligible overhead to the generation pipeline and can be seamlessly applied to any model. MLPM achieves state-of-the-art performance on diverse moderation benchmarks and demonstrates strong scalability across model families of various sizes. Moreover, we show that it integrates smoothly into end-to-end moderation pipelines and further improves response safety when combined with output moderation techniques. Overall, our work provides a practical and adaptable solution for safe, robust, and efficient LLM deployment.
Science Journals
arXiv:2606.02137v1 Announce Type: new
Abstract: In this work, we determine the generator polynomials for the Hermitian hulls and Hermitian sums of cyclic codes defined over the composite ring $\mathbb{F}_2 \times (\mathbb{F}_2 + v\mathbb{F}_2)$, where $v^2 = v$. Based on these structures, we develop quantum error-correcting (QEC) codes by applying the Hermitian dual version of Quantum Construction~X to the obtained Hermitian hulls and sums. Moreover, by employing matrix product code methods on linear complementary dual (LCD) codes defined over the same ring, we derive families of entanglement-assisted quantum error-correcting (EAQEC) codes.
arXiv:2602.07955v2 Announce Type: replace
Abstract: Crowd scenes captured by cameras at different locations vary greatly, and existing crowd models have limited generalization for unseen surveillance scenes. To improve the generalization of the model, we regard different surveillance scenes as different category scenes, and introduce few-shot learning to make the model adapt to the unseen surveillance scene that belongs to the given exemplar category scene. To this end, we propose to leverage local and global density characteristics to guide the model of crowd counting for unseen surveillance scenes. Specifically, to enable the model to adapt to the varying density variations in the target scene, we propose the multiple local density learner to learn multi prototypes which represent different density distributions in the support scene. Subsequently, these multiple local density similarity matrixes are encoded. And they are utilized to guide the model in a local way. To further adapt to the global density in the target scene, the global density features are extracted from the support image, then it is used to guide the model in a global way. Experiments on three surveillance datasets shows that proposed method can adapt to the unseen surveillance scene and outperform recent state-of-the-art methods in the few-shot crowd counting.
arXiv:2602.08236v2 Announce Type: replace
Abstract: Despite rapid progress in MLLMs, visual spatial reasoning remains unreliable when correct answers depend on how a scene would appear under unseen or alternative viewpoints. Recent work addresses this by augmenting reasoning with world models for visual imagination, but questions such as when imagination is actually necessary, how much of it is beneficial, and when it becomes harmful, remain poorly understood. In practice, indiscriminate imagination can increase computation and even degrade performance by introducing misleading evidence. In this work, we present an in-depth analysis of test-time visual imagination as a controllable resource for spatial reasoning. We first study when static visual evidence is sufficient, when imagination improves reasoning, and how excessive or unnecessary imagination affects accuracy and efficiency. To support this analysis, we then introduce AVIC, an adaptive test-time framework with world models that explicitly reasons about the sufficiency of current visual evidence before selectively invoking and scaling visual imagination. Finally, to further learn this gating and planning behavior without any annotation of when and how much to imagine, we introduce AVIC-R, which trains the policy via GRPO from QA-correctness rewards and penalties by imagination cost. Across spatial reasoning benchmarks (SAT, MMSI) and an embodied navigation benchmark (R2R), our results reveal clear scenarios where imagination is critical, marginal, or detrimental, and show that selective control can match or outperform fixed imagination strategies with substantially fewer world-model calls and language tokens. Our AVIC-R surpasses strong proprietary baselines including GPT-4o and GPT-4.1 while invoking the world model less often. Overall, our findings highlight the importance of analyzing and controlling test-time imagination for efficient and reliable spatial reasoning.
arXiv:2602.08585v2 Announce Type: replace
Abstract: Given the quadratic complexity of attention, KV cache eviction is vital to accelerate model inference. Current KV cache eviction methods typically rely on instantaneous heuristic metrics, implicitly assuming that score magnitudes are consistent proxies for importance across all heads. However, this overlooks the heterogeneity in predictive fidelity across attention heads. While certain heads prioritize the instantaneous contribution of tokens, others are dedicated to capturing long-horizon utility. In this paper, we propose that optimal budget allocation should be governed by the marginal utility in preserving long-term semantic information. Building on this insight, we propose LU-KV, a novel framework that formulates head-level budget allocation as a global combinatorial optimization problem to maximize the long-horizon marginal contribution of reserved tokens. To solve this non-convex problem, we employ a convex-hull relaxation and a marginal-utility-based greedy solver, achieving near-optimal solutions. Furthermore, we implement a data-driven offline profiling protocol to facilitate the practical deployment of LU-KV. Evaluations on LongBench and RULER benchmarks demonstrate that LU-KV reduces KV cache size by 80% with minimal performance degradation, while also decreasing inference latency and GPU memory footprint.
arXiv:2602.08646v3 Announce Type: replace
Abstract: We propose a gradient preconditioning method that makes reward-guided generation with one-step generative models both efficient and reliable. Test-time noise optimization can unlock substantially better reward-guided generations from pretrained generative models, but it is prone to reward hacking that degrades quality and is often too slow for practical use. We precondition reward gradients by projecting them onto a carefully designed white Gaussian noise feasible set, a compact spectral set with blockwise norm constraints that tightly captures the statistics and spatial uncorrelatedness of white Gaussian noise. This preconditioning reshapes each gradient update into a noise-aligned direction, driving faster and more effective reward ascent while preventing reward hacking. The projection is closed-form and matches the $O(N \log N)$ complexity of FFT, adding negligible overhead in practice. In experiments on FLUX with four reward models, our approach reaches a comparable Aesthetic Score using only 30% of the wall-clock time required by the state-of-the-art regularization-based method.
arXiv:2602.08689v2 Announce Type: replace
Abstract: Diffusion models generate samples through an iterative denoising process guided by a pretrained neural network. Once the denoiser is fixed, the sampling algorithm itself (noise schedules, guidance scales, stochasticity profiles) still requires careful tuning, a process typically carried out through costly empirical grid search. In this work, we introduce an inverse reinforcement learning framework for learning sampling strategies without retraining the denoiser. We formulate the diffusion sampling procedure as a discrete-time finite-horizon Markov Decision Process, where actions correspond to optional modifications of the sampling dynamics. To optimize action scheduling, we avoid defining an explicit reward function and instead directly match the target behavior expected from the sampler using policy gradient techniques. We provide experimental evidence that this approach matches fine-tuned samplers and comes at a modest cost compared to grid search: on ImageNet-64, a single training run replaces exhaustive search at up to 9x lower cost, with only 16% overhead at inference.
arXiv:2503.06520v3 Announce Type: replace
Abstract: Traditional methods for reasoning segmentation rely on supervised fine-tuning with categorical labels and simple descriptions, limiting its out-of-domain generalization and lacking explicit reasoning processes. To address these limitations, we propose Seg-Zero, a novel framework that demonstrates remarkable generalizability and derives explicit chain-of-thought reasoning through cognitive reinforcement. Seg-Zero introduces a decoupled architecture consisting of a reasoning model and a segmentation model. The reasoning model interprets user intentions, generates explicit reasoning chains, and produces positional prompts, which are subsequently used by the segmentation model to generate precious pixel-level masks. We design a sophisticated reward mechanism that integrates both format and accuracy rewards to effectively guide optimization directions. Trained exclusively via reinforcement learning with GRPO and without explicit reasoning data, Seg-Zero achieves robust zero-shot generalization and exhibits emergent test-time reasoning capabilities. Experiments show that Seg-Zero-7B achieves a zero-shot performance of 57.5 on the ReasonSeg benchmark, surpassing the prior LISA-7B by 18\%. This significant improvement highlights Seg-Zero's ability to generalize across domains while presenting an explicit reasoning process.
arXiv:2509.12795v3 Announce Type: replace
Abstract: In recent years, unmanned aerial vehicles (UAVs) have become increasingly popular in our daily lives and have attracted significant research interest in software engineering. At the same time, large language models (LLMs) have made notable advancements in language understanding, reasoning, and generation, making LLM applications in UAVs a promising research direction. However, existing studies have largely remained in preliminary exploration with a limited understanding of real-world practice, which causes an academia-industry gap and hinders the application of LLMs in UAVs. To address this, we conducted the first empirical study to investigate how LLMs support UAVs. To characterize common tasks and application scenarios of real-world UAV-LLM practices, we conducted a large-scale empirical study involving 997 research papers and 1,509 GitHub projects. The results classified nine common tasks (e.g., Natural Language Command Parsing) in four UAV workflows (e.g., Information Input) undertaken by LLMs in real-world UAV projects and revealed a large difference in the task distribution of research efforts and industry practices. To gain deeper insight into these differences and understand developers' perspectives on the application of LLMs in UAVs, we conducted a survey of practitioners, receiving 52 valid responses from 15 countries. The results revealed that while 40.4% of developers have attempted to apply LLMs to UAV tasks, 59.6% still face challenges integrating their UAV projects with advanced LLM capabilities. Their feedback attributes these challenges to five factors, including technological maturity, performance, safety, cost, and others, and provides practical implications for researchers and developers in conducting UAV-LLM practices.
arXiv:2602.08868v2 Announce Type: replace
Abstract: Time-series anomaly detection (TSAD) with multimodal large language models (MLLMs) is an emerging area, yet a persistent challenge remains: MLLMs rely on coarse time-series heuristics but struggle with multi-dimensional, detailed reasoning, which is vital for understanding complex time-series data. We present AnomSeer to address this by reinforcing the model to ground its reasoning in precise, structural details of time series, unifying anomaly classification, localization, and explanation. At its core, an expert chain-of-thought trace is generated to provide a verifiable, fine-grained reasoning from classical analyses (e.g., statistical measures, frequency transforms). Building on this, we propose a novel time-series grounded policy optimization (TimerPO) that incorporates two additional components beyond standard reinforcement learning: a time-series grounded advantage based on optimal transport and an orthogonal projection to ensure this auxiliary granular signal does not interfere with the primary detection objective. Across diverse anomaly scenarios, AnomSeer, with Qwen2.5-VL-3B/7B-Instruct, outperforms larger commercial baselines (e.g., GPT-4o) in classification and localization accuracy, particularly on point- and frequency-driven exceptions. Moreover, it produces plausible time-series reasoning traces that support its conclusions.
arXiv:2602.10014v3 Announce Type: replace
Abstract: Iterative self-improvement fine-tunes an autoregressive large language model (LLM) on reward-verified outputs generated by the LLM itself. In contrast to the empirical success of self-improvement, the theoretical foundation of this generative, iterative procedure in a practical, finite-sample setting remains limited. We make progress toward this goal by modeling each round of self-improvement as maximum-likelihood fine-tuning on a reward-filtered distribution and deriving finite-sample guarantees for the expected reward. Our analysis reveals an explicit feedback loop where better models accept more data per iteration, supporting sustained self-improvement while explaining eventual saturation of such improvement. Adopting a task-centric view by considering reasoning tasks with multiple difficulty levels, we further prove quantifiable conditions on model initialization, task difficulty, and sample budget where easy-to-hard curricula provably achieve better guarantees than training on fixed mixtures of tasks. Our analyses are validated through Monte-Carlo simulations and experiments spanning a synthetic graph-based reasoning task and multiple standard mathematical reasoning benchmarks.
arXiv:2602.10056v2 Announce Type: replace
Abstract: We introduce WildCat, a high-accuracy, low-cost approach to compressing the attention mechanism in neural networks. While attention is a staple of modern network architectures, it is also notoriously expensive to deploy due to resource requirements that scale quadratically with the input sequence length $n$. WildCat avoids these quadratic costs by only attending over a small weighted coreset. Crucially, we select the coreset using a fast but spectrally-accurate subsampling algorithm -- randomly pivoted Cholesky -- and weight the elements optimally to minimise reconstruction error. Remarkably, given bounded inputs, WildCat approximates exact attention with super-polynomial $O(n^{-\sqrt{\log(\log(n))}})$ error decay while running in near-linear $O(n^{1+o(1)})$ time. In contrast, prior practical approximations either lack error guarantees or require quadratic runtime to guarantee such high fidelity. We couple this advance with a GPU-optimized PyTorch implementation and a suite of benchmark experiments demonstrating the benefits of WildCat for image generation, image classification, and language model KV cache compression.
arXiv:2602.10120v2 Announce Type: replace
Abstract: We provide a numerical method for computing solutions to a free boundary problem arising from the equilibrium state of a floating drop. This numerical method is based on a Newton's method for the underlying nonlinear boundary value problems, and at each iterative step a Chebyshev spectral collocation method is employed. The problems considered here are those that can be described by using generating curves, and include problems in $\mathbb{R}^2$ and $\mathbb{R}^3$.
The resulting nine-dimensional space of physical parameters is explored, and examples are given that highlight the potential energy of centrally located drops, wall-bound drops, and asymmetrical configurations in $\mathbb{R}^2$. Non-uniqueness of solutions to the corresponding Euler-Lagrange equations is displayed, and also strong evidence of non-uniqueness of energy minimizers is given.
arXiv:2602.10623v2 Announce Type: replace
Abstract: Reward models learned from human preferences are central to aligning large language models (LLMs) via reinforcement learning from human feedback, yet they are often vulnerable to reward hacking due to noisy annotations and systematic biases such as response length or style. We propose Bayesian Non-Negative Reward Model (BNRM), a principled reward modeling framework that integrates non-negative factor analysis into Bradley-Terry (BT) preference model. BNRM represents rewards through a sparse, non-negative latent factor generative process that operates at two complementary levels: instance-specific latent variables induce disentangled reward representations, while sparsity over global latent factors acts as an implicit debiasing mechanism that suppresses spurious correlations. Together, this disentanglement-then-debiasing structure enables robust uncertainty-aware reward learning. To scale BNRM to modern LLMs, we develop an amortized variational inference network conditioned on deep model representations, allowing efficient end-to-end training. Extensive empirical results demonstrate that BNRM substantially mitigates reward over-optimization, improves robustness under distribution shifts, and yields more interpretable reward decompositions than strong baselines.
arXiv:2606.01234v1 Announce Type: cross
Abstract: The GDP of a country is modelled as the relative interaction between two agents - working hours, reflecting the social choice of a population, and Total Factor Productivity, reflecting the collective investment in productivity enhancers. It is shown that a Random Forest model can accu- rately predict the GDP from these two factors. The differences in the choices made by Germany and USA are analysed though Gini importance, SHAP plots and partial dependency. It is shown that the differences in the social structure of the countries are reflected in the relative contribution of working hours and productivity to the GDP.
arXiv:2509.20070v2 Announce Type: replace
Abstract: We present LLM Trainer, a fully automated pipeline that leverages the world knowledge of Large Language Models (LLMs) to transform a small number of human demonstrations (as few as one) into a large robot dataset for imitation learning. Our approach decomposes demonstration generation into two steps: (1) offline demonstration annotation that extracts keyframes, salient objects, and pose-object relations; and (2) online keypose retargeting that adapts those keyframes to a new scene, given an initial observation. Using these modified keypoints, our system warps the original demonstration to generate a new trajectory, which is then executed, and the resulting demo, if successful, is saved. Because the annotation is reusable across scenes, we use Thompson sampling to optimize the annotation, significantly improving generation success rate. We evaluate our method on a range of tasks, and find that our data annotation method consistently outperforms expert-engineered baselines. We further show an ensemble policy that combines the optimized LLM feed-forward plan with a learned feedback imitation learning controller. Finally, we demonstrate hardware feasibility on a Franka Emika Panda robot. For additional materials and demonstration videos, please see the project website: https://sites.google.com/andrew.cmu.edu/llm-trainer
arXiv:2509.21474v4 Announce Type: replace
Abstract: While diffusion language models (DLMs) have achieved competitive performance in text generation, improving their reasoning ability with reinforcement learning remains an active research area. Here, we introduce d2, a reasoning framework tailored for masked DLMs. Central to our framework is a new policy gradient algorithm that relies on accurate estimates of the sampling trajectory likelihoods. Because computing these likelihoods naively is computationally expensive for masked DLMs, we develop a family of estimators tailored to distinct model classes. For DLMs that support a sampling algorithm called any-order decoding, we propose d2-AnyOrder, which achieves exact trajectory likelihood with a single model pass. Through an empirical study of widely used DLMs, we show that any-order decoding is not universally supported in practice. For standard masked diffusion models, we propose d2-StepMerge, which approximates the trajectory likelihood, trading off compute for approximation accuracy in an analytically tractable manner. Empirically, d2 significantly outperforms widely-used RL baselines when applied to popular DLMs, and sets a new state-of-the-art performance for DLMs on logical reasoning tasks (Countdown and Sudoku) and math reasoning benchmarks (GSM8K and MATH500). We provide the code along with a blog post on the project page: https://guanghanwang.com/d2
arXiv:2503.11832v5 Announce Type: replace
Abstract: Recent vision language models (VLMs) have made remarkable strides in generative modeling with multimodal inputs, particularly text and images. However, their susceptibility to generating harmful content when exposed to unsafe queries raises critical safety concerns. While current alignment strategies primarily rely on supervised safety fine-tuning with curated datasets, we identify a fundamental limitation we call the ''safety mirage'', where supervised fine-tuning inadvertently reinforces spurious correlations between superficial textual patterns and safety responses, rather than fostering deep, intrinsic mitigation of harm. We show that these spurious correlations leave fine-tuned VLMs vulnerable even to a simple one-word modification-based attack, where substituting a single word in text queries with a spurious correlation-inducing alternative can effectively bypass safeguards. Additionally, these correlations contribute to the over-prudence, causing fine-tuned VLMs to refuse benign queries unnecessarily. To address these issues, we show machine unlearning (MU) as a powerful alternative to supervised safety fine-tuning, as it avoids biased feature-label mappings and directly removes harmful knowledge from VLMs while preserving their general capabilities. Extensive evaluations across safety benchmarks show that under MU-based alignment reduces the attack success rate by up to 60.27% and cuts unnecessary rejections by over 84.20%. WARNING: There exist AI generations that may be offensive in nature.
arXiv:2506.09220v2 Announce Type: replace
Abstract: A new generation of AI personal assistants reached consumers in 2023-2024 amid sweeping claims about anticipatory, agentic intelligence. Wearables such as the Rabbit R1 and Humane AI Pin, and subscription services such as Ohai and Docus, promised to learn users' routines and complete tasks across digital platforms. Drawing on semi-structured interviews with nine early adopters, this article asks how users make sense of these systems when the imaginary of an autonomous "second self" meets the recalcitrance of actual devices. Extending uncertainty reduction theory, we specify three forms of uncertainty in initial encounters: functional (what can it do?), relational (how do I get it to do it?), and metaphysical (what is it to me, and what should it remember?). We find that hype continues the pre-domestication of voice assistants; that the most satisfying uses are user-curated constellations of narrow tools rather than standalone "second selves."
arXiv:2606.00393v1 Announce Type: cross
Abstract: Geometric distortion in prostate diffusion-weighted imaging (DWI) can impair lesion localization and reduce the reliability of MRI-based clinical assessment. We propose AutoIQ, an ensemble machine learning framework for automatic quantification and classification of DWI geometric distortion severity. A total of 140 retrospective prostate biparametric MRI examinations were analyzed, including 33 scans with severe distortion requiring repeat acquisition and 107 scans with acceptable distortion based on expert radiologist assessment. AutoIQ combines two complementary distortion quantification strategies: a segmentation-based method measuring prostate boundary mismatch between T2-weighted imaging (T2WI) and DWI, and a registration-based method estimating deformation magnitude after DWI-to-T2WI alignment. The resulting distortion scores were used to train individual classifiers and a logistic-regression ensemble model. Both computational methods significantly differentiated severe from acceptable distortion cases (p < 0.001). On an independent test set, the ensemble model achieved an accuracy of 0.95, F1-score of 0.93, and AUC of 0.98, outperforming individual models. These results suggest that AutoIQ can provide automated, quantitative quality assessment for prostate DWI and may help identify scans that require repeat acquisition.
arXiv:2606.00403v1 Announce Type: cross
Abstract: Predicting how crystal structure influences high-temperature stability remains a key challenge in materials modelling and design. Silicon carbide (SiC), one of the most thermally and chemically stable materials known, provides an ideal system for studying this problem because its many polytypes preserve similar local tetrahedral bonding while differing in long-range stacking geometry. Here, we combine phase-coexistence machine-learning molecular dynamics with finite-temperature phonon analysis, enabled by a fine-tuned MACE interatomic potential that accurately describes crystalline, high-temperature, and disordered configurations across multiple SiC polytypes. We identify a clear relative stability ordering, 3C > 2H > 9R, reflected consistently in structural disordering, interlayer sliding, and finite-temperature phonon spectra. Across all polytypes, melting initiates through the formation of short C-C contacts and carbon-rich local regions, followed by a progressive loss of tetrahedral Si-C connectivity. The reduced stability of the long-period 9R polytype is traced to low-frequency transverse-acoustic shear modes associated with relative bilayer sliding, which are already present in the 0 K phonon spectra and soften further at high temperature. These modes generate larger lateral bilayer displacements, linking enhanced interlayer sliding to local chemical disordering and ultimately melting. More broadly, our results show that high-temperature stability in polytypic covalent materials is governed not only by local bond strength, but also by stacking-dependent transverse dynamics.
arXiv:2606.00409v1 Announce Type: cross
Abstract: Unconventional antiferromagnets combine zero net magnetization with spin-split electronic bands, offering a distinct, important platform for spintronics. Their discovery, however, has so far depended largely on case-by-case studies and on a limited number of compounds with experimentally resolved magnetic structures. Here, we overcome these bottlenecks by resolving magnetic ground states across a broad materials database. We narrow down 37163 magnets from the Materials Project to 189 collinear antiferromagnets by combining physics-informed prescreening, high-throughput exchange calculations and Luttinger-Tisza analysis. Among these, symmetry analysis identifies 36 altermagnets and 11 Luttinger-compensated ferrimagnets (LCFs), including 22 altermagnets and 9 LCFs that have not been reported previously. The identified unconventional antiferromagnets can support nonrelativistic spin Hall effects and doping-tunable spin transport with switchable polarization and giant anisotropy. Our framework converts broad structural databases into a curated, symmetry-classified set of experimentally testable compensated spin-split magnets, establishing a scalable route for the efficient discovery of functional antiferromagnets.
arXiv:2606.00413v1 Announce Type: cross
Abstract: Sufficient dimension reduction (SDR) makes high-dimensional regression tractable by projecting the covariates onto a low-dimensional subspace that preserves the conditional mean of the response. Existing gradient-based estimators either operate in the ambient space and suffer from the curse of dimensionality, or localize in the reduced space at a per-outer-iteration cost at least quadratic in the sample size. We show that minimizers of the population Minimum Average Variance Estimation (MAVE) risk approximate the same Grassmannian target as the Outer Product of Gradients (OPG), and recast the empirical criterion as a smooth maximization on the Stiefel manifold with closed-form Riemannian gradient. The resulting algorithm, SMAVE, combines sparse projected-space nearest-neighbor localization with Riemannian stochastic gradient ascent. A simplified version comes with almost-sure convergence and a non-asymptotic rate matching the standard non-convex stochastic first-order scaling. Empirically, SMAVE matches or improves on RMAVE's synthetic subspace recovery at moderate-to-high ambient dimension, and on four real datasets it uniformly improves over OPG and is competitive with or outperforms RMAVE at orders of magnitude lower runtime.
arXiv:2602.11554v3 Announce Type: replace
Abstract: How far can 3D object detection go using 4D radar alone? Despite offering weather-robust and velocity-aware sensing for autonomous perception, modern 4D radar still yields sparse, noisy, and unstable point clouds, limiting radar-only 3D detection. We present HyperDet, a detector-agnostic framework that constructs task-aware hyper 4D radar point clouds before detection. HyperDet first refines short-window surround-view radar observations through spatio-temporal accumulation, cross-sensor validation, and Doppler-guided motion compensation, improving return reliability and temporal coherence. It then performs foreground generative enhancement using LiDAR-guided pseudo-radar supervision available only during training, enriching object geometry while preserving measured radar background and radar-native attributes. During detector training, radar-aware object-level augmentation further preserves Doppler consistency under geometric relocation. At inference time, HyperDet requires radar input alone and can be directly paired with standard 3D detectors. Experiments on two public surround-view 4D radar datasets demonstrate consistent improvements over raw radar inputs across standard 3D detectors, validating input-level radar enhancement as an effective approach to radar-only 3D detection.
arXiv:2602.13937v2 Announce Type: replace
Abstract: Automated Machine Learning (AutoML) has improved access to machine learning, yet existing techniques often remain limited in flexibility, transparency, and execution reliability. Code-driven AutoML offers a promising direction by synthesizing executable code for preprocessing, model training, and evaluation. However, current LLM-based approaches frequently generate code that is plausible in text yet brittle in execution, insufficiently grounded in the actual dataset, or restricted to narrow solution paths. In this paper, we introduce iML, a multi-agent code-driven AutoML framework designed around three requirements: executability, problem grounding, and broad exploration of valid solutions. iML first analyzes the task and profiles the data, then synthesizes a structured blueprint that guides modular code generation across multiple implementation tracks, including traditional ML,pretrained adaptation, and custom neural architectures. To improve reliability, iML enforces interface checking, dynamic execution, and iterative debugging during integration. We evaluate iML on MLE-BENCH and the newly introduced iML-BENCH, covering diverse Kaggle-style tasks. On MLE-BENCH, iML attains a 90% valid submission rate and a 45% medal rate, and an APS of 0.82, improving the average standardized performance score (APS) over the LLM-based baselines by 52%-273%. On iML-BENCH, it achieves the highest APS and demonstrates robust performance even when task descriptions are substantially stripped. These results establish iML as a reliable and competitive framework for code-driven AutoML.