arXiv subjects

Yinghui He

Publications and source records attributed to Yinghui He.

At least 19 recordsLinked to original sources

arXiv2026

CritICL: Inference-Time Weak-to-Strong Generalization from Small Language Model Failure Modes

Recent advances in inference-time scaling have significantly improved the reasoning performance of large language models (LLMs). However, these methods typically rely on repeated generation or external verification. To address this limitation, we introduce CritICL, a novel inference-time framework that improves reasoning while maintaining high efficiency. Our key insight is that LLM failure modes exhibit structured patterns across model scales within the same family. Instead of treating failures as undesirable outputs, CritICL leverages them as a source of guidance. Specifically, we utilize failure modes derived from weaker models and incorporate them into inference through critique-based in-context examples. We propose two variants: CritICL-dynamic, which adaptively predicts input-specific failure modes and retrieves critiques, and CritICL-static, which uses a global failure mode profile to provide stable guidance. Experimental results show that CritICL consistently outperforms standard in-context learning and achieves performance competitive with or superior to test-time scaling methods, while requiring significantly fewer generations and lower token cost. Code available at: https://github.com/umwyf/CRITICL

cs.CL↗

arXiv2026

Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning

Long-horizon reasoning in recent LLMs demands that the model switch between distinct skills inside a reasoning chain, such as first doing a math derivation, then using the result to plan a schedule. We call such problems cross-skill long-horizon tasks: multi-step tasks whose steps require different reasoning skills and depend on earlier outputs. Existing benchmarks often evaluate individual skills, lacking a principled way to measure how well a model switches between skills. We address this gap from both the evaluation and training sides. We introduce Skill Entropy, a measure of the difficulty of switching from one skill to another. We then propose Skill^2-Bench, a benchmark of cross-skill long-horizon tasks built over 558 skills across 9 verifiable and open-ended domains. Each task is assigned a task-level skill-entropy score and grouped into three difficulty levels. Evaluating 8 frontier and 4 open-source models on Skill^2-Bench reveals a skill-switching gap: accuracy decreases on higher-entropy tasks. We then turn skill entropy from a benchmark scale into a training signal. We propose Skill-Entropy RL, an RL framework where the model predicts not only the answer at each step but also the skill used to produce it. The reward combines step-level correctness with a skill-entropy reward that measures the alignment between the model-predicted skill sequence and the gold skill sequence. On Qwen3-4B-Instruct and Qwen3-1.7B, Skill-Entropy RL improves the Skill^2-Bench score from 34.4% to 68.4% and from 14.6% to 40.1%, respectively, outperforming competitive baselines. The same pipeline can be applied to off-the-shelf training data such as OpenR1-Math, indicating that skill entropy is a reusable training signal. Code available at: https://github.com/Gen-Verse/Skill-Entropy-RL

cs.CL↗

arXiv2026

T^2MLR: Transformer with Temporal Middle-Layer Recurrence

Transformer reasoning is limited by autoregressive decoding, which repeat edly compresses rich hidden computation through token space and makes it difficult for intermediate reasoning states to persist across time. We in troduce Transformers with Temporal Middle-Layer Recurrence (T2MLR), a transformers-based latent reasoning architecture that fuses a cached middle layer representation from the previous token directly into an earlier layer of the current token position, enabling abstract intermediate computation to persist across decoding steps with little inference overhead. Across natural-language pretraining and multi-hop reasoning finetuning, T2MLR consistently outperforms data- and parameter-matched Transformer base lines. Moreover, applying recurrence to only a localized middle-layer block (as little as 20% of the network) often outperforms full-layer recurrence. Im portantly, T2MLR does not require pretraining from scratch: retrofitting the recurrent pathway into an existing pretrained 1.7B Transformer and briefly finetuning substantially improves math reasoning, lowering the barrier to practical adoption. These results suggest that effective latent reasoning in Transformers does not require looping over all layers as in previous works, but can instead emerge more strongly from targeted middle-layer recurrence.

cs.CL↗

arXiv2026

Rethinking On-Policy Self-Distillation for Thinking Models

Self-distillation is a promising recipe for self-improvement in language models. In this setting, a model can serve as its own teacher when given privileged information, such as a solution to a math problem. This seems especially appealing for thinking models, which can use test-time reasoning to absorb the privileged information. Surprisingly, we show that privileged self-distillation degrades thinking models on long reasoning traces: across five Qwen3 and OLMo thinking models evaluated on AIME24, AIME25, and HMMT25, privileged-context distillation causes a relative drop of up to 17% in avg@16 accuracy. The degradation scales with the amount of privileged context withheld from the student and is most pronounced at long rollout budgets, where thinking models otherwise obtain their largest gains. This failure mode is not specific to self-distillation: on-policy distillation (OPD) improves thinking models, but privileged OPD reverses these gains. Our diagnostics link this failure mode to how privileged teacher context reshapes learning at high-entropy forking positions, where multiple continuations remain plausible and may lead to different reasoning paths. Privileged context lowers fork rates in thinking-model rollouts but not in instruction-model rollouts. This leads to an interesting dichotomy, where privileged context can help instruction-tuned models but hurts stronger thinking models. The effect is visible when the student begins a self-correction branch, where privileged OPD penalizes sampled reconsideration tokens that vanilla OPD supports. Thinking models trained with a privileged teacher produce fewer verification, backtracking, and hedging markers, even after length normalization. These findings indicate that self-distillation for strong thinking models requires attention to token-level signal, especially around correction and reasoning steps.

cs.AI↗

arXiv2026

Task-Oriented Multimodal Edge Intelligence via Integrated Sensing-Communication-Computation

Integrated sensing, communication, and computation (ISCC) has recently emerged as a unified framework for enabling edge intelligence. However, existing ISCC designs predominantly rely on single-modal sensing, which is inherently vulnerable to occlusions, environmental uncertainties, and modality-specific failures, leading to degraded robustness in real-world deployments. This motivates the need for multi-modal ISCC, yet its design remains insufficiently explored. Compared with the single-modal case, multi-modal ISCC is more challenging because heterogeneous modalities enlarge data dimensionality and tighten communication/computation/energy budgets, while inter-modal correlations further complicate performance characterization. To address these challenges, we propose a task-oriented multi-modal ISCC framework that integrates device-side feature extraction with edge-side joint multi-modal inference. A central component of our approach is the maximal coding rate reduction (MCR^2) criterion, which enables each device to learn compact and discriminative task-relevant features, offering clear advantages over conventional cross-entropy-based extractors. We further leverage MCR^2 as a principled metric for edge-side sensing evaluation. On this basis, we formulate a sensing accuracy maximization problem under delay and resource constraints and develop an efficient block coordinate descent (BCD) algorithm after transforming the problem into a more tractable equivalent form. Focusing on a human activity recognition task, we conduct extensive experiments on publicly available datasets to evaluate the performance of the proposed ISCC framework. The results demonstrate that our approach consistently outperforms three baseline schemes under limited resource conditions.

eess.SP↗

arXiv2026

Edge-Assisted Multimodal UAV Localization with Resource-Efficient Compression and Robust Fusion

The unmanned aerial vehicles (UAVs) will play an important role in the future urban transportation systems. This requires designing robust localization schemes especially for non-cooperative UAVs that do not share any information about their movements. This paper designs a multimodal UAV localization framework which utilizes camera, LiDAR and radar sensing modalities. The underlying data processing and the subsequent inference of the UAV location are distributed among the sensing nodes and the edge server attached to the base station. The proposed UAV localization framework addresses three key challenges. First, the sensing nodes have limited computing and communication resources, and they contain only single modality sensors. Second, the multimodal data differ greatly in the sampling rates, time alignment and the encodings. Third, the changes in the environment and the hardware failures cause the modal data to degrade, or to be completely missing. The proposed localization framework utilizes several data processing modules including a information-bottleneck (IB)-based compression module that extracts the most relevant features from each modality, a time-encoding alignment module that provides the unified representation in a shared latent space, a multimodal fusion module that accounts for the degraded and missing data, and a Mamba-based regression module that predicts the present UAV location. The experiments involving a real-world dataset demonstrate that the proposed framework accurately and reliably obtains the UAV location while outperforming other existing frameworks.

eess.SP↗

arXiv2026

Toward Efficient Sensing in Multi-Device ISCC by Removing Frequency Domain Redundancy

Integrated sensing, communication, and computation (ISCC) is envisioned as a key enabler for intelligent services in future wireless networks. However, in multi-device ISCC systems, directly offloading full orthogonal frequency division multiplexing (OFDM) sensing data to the edge may incur excessive overhead, thereby limiting sensing performance under practical resource constraints. In this paper, we propose a subcarrier selection-based sensing framework for multi-device ISCC systems, where frequency-domain redundancy in OFDM sensing data is removed during local preprocessing to reduce sensing data transmission and processing overhead. Based on the proposed framework, we establish analytical models for sensing accuracy, delay, and energy consumption, and formulate a sensing accuracy maximization problem under practical resource constraints. To solve this problem, we develop an alternating direction method of multipliers (ADMM)-based algorithm. Experiments on commodity wireless devices validate the effectiveness of the proposed framework and show that it consistently outperforms three baseline schemes under various resource constraints.

eess.SP↗

arXiv2026

Transferability for General Reasoning: An Automated Curriculum for Multi-Domain RLVR

Reinforcement learning with verifiable rewards (RLVR) has been extended from single-domain training to multi-domain reasoning suites spanning mathematics, programming, and science. However, the training curriculum (how often each domain is sampled) is typically fixed or hand-tuned, even though reasoning skills transfer unevenly across domains. Existing learnability-based curricula adapt to where the policy is currently improving, but are blind to whether a gradient step on the selected domain benefits the remaining domains. In this paper, we propose Transfer-Aware Curriculum (TAC), a bandit-style online curriculum that prioritizes domains whose updates broadly benefit the rest of the training suite. TAC repurposes signals already produced by RL training: per-domain advantages capture local learnability, and projected gradients, taken from the GRPO step being computed, estimate cross-domain transferability via gradient-geometry alignment, at negligible cost (<1% wall-clock overhead). Across a six-domain reasoning suite, TAC achieves the best macro-averaged accuracy on both Qwen3-1.7B and Llama3.2-3B, outperforming proportional random sampling, a hand-designed schedule, and a learnability-only bandit, and improving over the last of these by up to 2.8 points (10% relative). Ablations show performance degrades sharply when the transferability term is removed, and TAC remains robust on imbalanced training mixtures where learnability-only curricula over-commit to dominant domains. Our findings establish cross-domain transferability as a key signal for curriculum design in multi-domain RLVR.

cs.AI↗

arXiv2026

MixSD: Mixed Contextual Self-Distillation for Knowledge Injection

Supervised fine-tuning (SFT) is widely used to inject new knowledge into language models, but it often degrades pretrained capabilities such as reasoning and general-domain performance. We argue this forgetting arises because fine-tuning targets from humans or external systems diverge from the model's autoregressive distribution, forcing the optimizer to imitate low-probability token sequences. To address this problem, we propose MixSD, a simple external-teacher-free method for distribution-aligned knowledge injection. Instead of training on fixed targets, MixSD constructs supervision dynamically by mixing tokens from two conditionals of the base model itself: an expert conditional that observes the injected fact in context, and a naive conditional that reflects the model's original prior. The resulting supervision sequences preserve the factual learning signal while remaining substantially closer to the base model's distribution. We evaluate MixSD on two synthetic corpora that we construct to study factual recall and arithmetic function acquisition in a controlled setting, together with established benchmarks for open-domain factual question answering and knowledge editing. Across multiple model scales and settings, MixSD consistently achieves a better memorization-retention trade-off compared to SFT and on-policy self distillation baselines, retaining up to 100% of the base model's held-out capability while maintaining near-perfect training accuracy, whereas standard SFT retains as little as 1%. We further show that MixSD produces substantially lower-NLL supervision targets under the base model and reduces harmful movement along Fisher-sensitive parameter directions. These results suggest that aligning supervision with the model's native generation distribution is a simple and effective principle for knowledge injection that mitigates catastrophic forgetting.

cs.CL↗

arXiv2026

High-Fidelity and Location-Robust Respiratory Waveform Monitoring with Single-Antenna WiFi

In recent years, WiFi sensing has been recognized as a promising technology to bring respiratory monitoring into everyday homes, thanks to its contactless nature and ubiquitous availability. However, existing WiFi-based respiratory monitoring systems still fall short of deployment-oriented performance: they suffer from restrained hardware scalability, limited accuracy, and are highly sensitive to user location. To overcome these limitations and push WiFi sensing towards clinically meaningful precision, we propose RespirFi, a novel system that robustly delivers high-fidelity respiratory waveforms with WiFi Channel State Information (CSI), thereby enabling accurate estimation of key physiological biomarkers. At the core of RespirFi is a theoretical human reflection model, through which we perform an in-depth characterization of how CSI variations are shaped by both subcarrier frequency and spatial user location. Guided by these insights, we develop a location-robust waveform construction method that adaptively selects high quality subcarriers and aligns their waveform trends, ensuring accurate waveform recovery. Furthermore, we propose a breathing phase identification method that leverages inter-subcarrier CSI differences to reliably distinguish inhalation from exhalation. We implement RespirFi over commodity WiFi devices, and extensive experiments demonstrate that it outperforms state-of-the-art approaches across a wide range of clinically relevant respiratory metrics.

eess.SP↗

arXiv2026

Self-Distillation Zero: Self-Revision Turns Binary Rewards into Dense Supervision

Current post-training methods in verifiable settings fall into two categories. Reinforcement learning (RLVR) relies on binary rewards, which are broadly applicable and powerful, but provide only sparse supervision during training. Distillation provides dense token-level supervision, typically obtained from an external teacher or using high-quality demonstrations. Collecting such supervision can be costly or unavailable. We propose Self-Distillation Zero (SD-Zero), a method that is substantially more training sample-efficient than RL and does not require an external teacher or high-quality demonstrations. SD-Zero trains a single model to play two roles: a Generator, which produces an initial response, and a Reviser, which conditions on that response and its binary reward to produce an improved response. We then perform on-policy self-distillation to distill the reviser into the generator, using the reviser's token distributions conditioned on the generator's response and its reward as supervision. In effect, SD-Zero trains the model to transform binary rewards into dense token-level self-supervision. On math and code reasoning benchmarks with Qwen3-4B-Instruct and Olmo-3-7B-Instruct, SD-Zero improves performance by at least 10% over the base models and outperforms strong baselines, including Rejection Fine-Tuning (RFT), GRPO, and Self-Distillation Fine-Tuning (SDFT), under the same question set and training sample budget. Extensive ablation studies show two novel characteristics of our proposed algorithm: (a) token-level self-localization, where the reviser can identify the key tokens that need to be revised in the generator's response based on reward, and (b) iterative self-evolution, where the improving ability to revise answers can be distilled back into generation performance with regular teacher synchronization. Code: https://github.com/princeton-pli/Self-Distillation-Zero.

cs.CL↗

arXiv2026

Path to Diversity: A Primer on ISAC-izing Commodity Wi-Fi for Practical Deployments

Integrated Sensing and Communication (ISAC) has emerged as a key paradigm in next-generation wireless networks. While the ubiquity and low cost of commodity Wi-Fi make it an ideal platform for wide-scale sensing, it is the continuous evolution of Wi-Fi standards-towards higher frequency bands, wider bandwidths, and larger antenna arrays-that fundamentally unlocks the physical resources required for high-performance ISAC. To structure this rapidly expanding field, numerous surveys have appeared. However, prevailing literature predominantly adopts a top-down perspective, emphasizing upper-layer applications or deep learning models while treating the physical layer as an opaque abstraction. Consequently, these works often fail to touch the bottom layer of signal formation and lack technical guidance on overcoming the physical barriers that constrain sensing performance. To bridge this gap, this tutorial takes a bottom-up approach, systematically analyzing the sensing gains brought by Wi-Fi advancements through the lens of physical-layer diversity. We organize the framework around four orthogonal dimensions: i) Temporal Diversity addresses synchronization gaps to enable absolute ranging; ii) Frequency Diversity expands the effective bandwidth to sharpen range resolution; iii) Link Diversity leverages distributed topologies and digital feedback to achieve ubiquitous observability; and iv) Spatial Diversity utilizes multi-antenna arrays to combine passive angular discrimination with active directional control. Collectively, these orthogonal dimensions resolve fundamental ambiguities in time, range, and space, bridging physical capabilities with challenging sensing diversities. By synthesizing these dimensions, this tutorial provides a comprehensive guide for "ISAC-izing" commodity Wi-Fi, paving the way for future standardization and robust deployment.

cs.NI↗

arXiv2026

MobiDiary: Autoregressive Action Captioning with Wearable Devices and Wireless Signals

Human Activity Recognition (HAR) in smart homes is critical for health monitoring and assistive living. While vision-based systems are common, they face privacy concerns and environmental limitations (e.g., occlusion). In this work, we present MobiDiary, a framework that generates natural language descriptions of daily activities directly from heterogeneous physical signals (specifically IMU and Wi-Fi). Unlike conventional approaches that restrict outputs to pre-defined labels, MobiDiary produces expressive, human-readable summaries. To bridge the semantic gap between continuous, noisy physical signals and discrete linguistic descriptions, we propose a unified sensor encoder. Instead of relying on modality-specific engineering, we exploit the shared inductive biases of motion-induced signals--where both inertial and wireless data reflect underlying kinematic dynamics. Specifically, our encoder utilizes a patch-based mechanism to capture local temporal correlations and integrates heterogeneous placement embedding to unify spatial contexts across different sensors. These unified signal tokens are then fed into a Transformer-based decoder, which employs an autoregressive mechanism to generate coherent action descriptions word-by-word. We comprehensively evaluate our approach on multiple public benchmarks (XRF V2, UWash, and WiFiTAD). Experimental results demonstrate that MobiDiary effectively generalizes across modalities, achieving state-of-the-art performance on captioning metrics (e.g., BLEU@4, CIDEr, RMC) and outperforming specialized baselines in continuous action understanding.

cs.CV↗

arXiv2026

Invisible Walls: Privacy-Preserving ISAC Empowered by Reconfigurable Intelligent Surfaces

The environmental and target-related information inherently carried in wireless signals, such as channel state information (CSI), has brought increasing attention to integrated sensing and communication (ISAC). However, it also raises pressing concerns about privacy leakage through eavesdropping. While existing efforts have attempted to mitigate this issue, they either fail to account for the needs of legitimate communication and sensing or rely on hardware with high cost. To overcome these limitations, we propose PrivISAC, a plug-and-play, low-cost solution that leverages reconfigurable intelligent surface (RIS) to protect user privacy while preserving ISAC performance. At its core, PrivISAC constructs a small set of RIS configurations from two beamforming vectors per RIS row and randomly activates one configuration per time slot to perturb CSI and mask sensitive sensing information from eavesdroppers. The two vectors are designed to maintain similar communication-direction responses while creating distinct sensing-direction responses, thereby preserving transmission quality and masking sensitive CSI from eavesdroppers. To enable legitimate sensing, we introduce a time-domain masking and demasking method that maps CSI samples to their RIS configurations and removes configuration-induced discrepancies. We implement PrivISAC on commodity wireless devices and conduct extensive experiments. Results show that PrivISAC provides effective privacy protection while preserving high-quality communication and sensing performance for legitimate receivers.

eess.SP↗

arXiv2025

DeepGuard: Defending Deep Joint Source-Channel Coding Against Eavesdropping at Physical-Layer

Deep joint source-channel coding (DeepJSCC) has emerged as a promising paradigm for efficient and robust information transmission. However, its intrinsic characteristics also pose new security challenges, notably an increased vulnerability to eavesdropping attacks. Existing studies on defending against eavesdropping attacks in DeepJSCC, while demonstrating certain effectiveness, often incur considerable computational overhead or introduce performance trade-offs that may adversely affect legitimate users. In this paper, we present DeepGuard, to the best of our knowledge, the first physical-layer defense framework for DeepJSCC against eavesdropping attacks, validated through over-the-air experiments using software-defined radios (SDRs). Considering that existing eavesdropping attacks against DeepJSCC are limited to simulation under ideal channels, we take a step further by identifying and implementing four representative types of attacks under various configurations in orthogonal frequency-division multiplexing systems. These attacks are evaluated over-the-air under diverse scenarios, allowing us to comprehensively characterize the real-world threat landscape. To mitigate these threats, DeepGuard introduces a novel preamble perturbation mechanism that modifies the preamble shared only between legitimate transceivers. To realize it, we first conduct a theoretical analysis of the perturbation's impact on the signals intercepted by the eavesdropper. Building upon this, we develop an end-to-end perturbation optimization algorithm that significantly degrades eavesdropping performance while preserving reliable communication for legitimate users. We prototype DeepGuard using SDRs and conduct extensive over-the-air experiments in practical scenarios. Extensive experiments demonstrate that DeepGuard effectively mitigates eavesdropping threats.

eess.SP↗

arXiv2025

Meta-SimGNN: Adaptive and Robust WiFi Localization Across Dynamic Configurations and Diverse Scenarios

To promote the practicality of deep learning-based localization, existing studies aim to address the issue of scenario dependence through meta-learning. However, these studies primarily focus on variations in environmental layouts while overlooking the impact of changes in device configurations, such as bandwidth, the number of access points (APs), and the number of antennas used. Unlike environmental changes, variations in device configurations affect the dimensionality of channel state information (CSI), thereby compromising neural network usability. To address this issue, we propose Meta-SimGNN, a novel WiFi localization system that integrates graph neural networks with meta-learning to improve localization generalization and robustness. First, we introduce a fine-grained CSI graph construction scheme, where each AP is treated as a graph node, allowing for adaptability to changes in the number of APs. To structure the features of each node, we propose an amplitude-phase fusion method and a feature extraction method. The former utilizes both amplitude and phase to construct CSI images, enhancing data reliability, while the latter extracts dimension-consistent features to address variations in bandwidth and the number of antennas. Second, a similarity-guided meta-learning strategy is developed to enhance adaptability in diverse scenarios. The initial model parameters for the fine-tuning stage are determined by comparing the similarity between the new scenario and historical scenarios, facilitating rapid adaptation of the model to the new localization scenario. Extensive experimental results over commodity WiFi devices in different scenarios show that Meta-SimGNN outperforms the baseline methods in terms of localization generalization and accuracy.

cs.LG↗

arXiv2025

On the Impossibility of Retrain Equivalence in Machine Unlearning

Machine unlearning seeks to selectively remove the "influence" of specific training data on a model's outputs. The ideal goal is Retrain Equivalence--behavior identical to a model trained from scratch on only the retained data. This goal was formulated for models trained on i.i.d. data batches, but modern pipelines often involve multi-stage training, with each stage having a distinct data distribution and objective. Examples include LLM fine-tuning for alignment, reasoning ability, etc. Our study shows via theory and experiments that this shift to multi-stage training introduces a fundamental barrier for machine unlearning. The theory indicates that the outcome of local unlearning--methods that only use gradients computed on the forget set--is path-dependent. That is, a model's behavior during unlearning is influenced by the order of its training stages during learning, making it impossible for path-oblivious algorithms to universally achieve Retrain Equivalence. We empirically demonstrate the same phenomenon in LLM post-training across Llama and Qwen models (1B to 14B) with gradient ascent, NPO, and SimNPO local unlearning algorithms. Models fine-tuned via different orderings of identical training stages diverge in behavior during unlearning, with the degradation in GSM8K accuracy after unlearning varying by over 20% across paths. We also observe that some learning paths consistently produce models that unlearn slowly. During unlearning, whether the probability mass gets squeezed into paraphrasing or alternative concepts is also path-dependent. These results consistently show that Retrain Equivalence is an ill-posed target for local unlearning algorithms, so long as the target models are trained in stages. In situations where access to models' training histories is hard, the current work calls for rethinking the definition and desiderata of machine unlearning.

cs.LG↗

arXiv2025

Skill-Targeted Adaptive Training

Language models often show little to no improvement (i.e., "saturation") when trained via vanilla supervised fine-tuning (SFT) on data similar to what they saw in their training set (e.g., MATH). We introduce a new fine-tuning strategy, STAT, to train such a student model by using the metacognition ability of a stronger large language model (LLM) as the teacher. The teacher uses the task dataset to create a list of skills needed for the task, and then labels each data point with its required skills (Didolkar et al., 2024). By monitoring the student's answers, the teacher creates a Missing-Skill-Profile for the student, tracking how often they failed to apply each skill in their responses. We use this idea to build a modified training set in one of two ways. In STAT-Sel, the teacher uses an existing set of training examples but adaptively reweights them according to the Missing-Skill-Profile. In STAT-Syn, the teacher synthesizes additional examples involving missing skills. Across extensive experiments on Llama and Qwen models, our methods yield improvements of up to 7.5% on MATH, whereas SFT provides only limited gains. Furthermore, STAT enhances performance on out-of-distribution benchmarks (e.g., AIME24/25, AMC23, etc.) by an average of 4.6%. Crucially, we find that STAT is complementary to RL via GRPO (Shao et al., 2024): after the model is improved using STAT to address skill gaps, GRPO continues to add further gains. We conclude that skill-targeted adaptive training should broadly improve current training pipelines. Our code is available at: https://github.com/princeton-pli/STAT.

cs.LG↗