Publications

Learning Where to Constrain: Meta-Learned Trust Allocation for Mathematical Reasoning
Long mathematical solutions expose a weakness in uniform policy-update constraints. Early token changes affect a long suffix, while late cha… (see more)nges affect fewer future states. Prefix divergence-based methods like CPPO encode this asymmetry with a fixed position schedule and prefix budget. We introduce Meta-TA, a bilevel method for learning both choices from post-update performance on disjoint prompts. A direct one-step design fails when behavior and current policies match because the constraint starts inactive. We prove the resulting hypergradient equals zero. A detached warm-up creates controlled policy drift. A soft-gated step then supplies the meta-gradient. On Qwen3-8B-Base, five frozen-retraining seeds raise mean AIME24/25/26 Avg@16 from 29.01 for CPPO to 31.95 for Meta-TA with closely matched realized KL. At a 16k response cap, Meta-TA lowers normalized reasoning effective horizon from 0.475 to 0.430 and importance-ratio variance from 0.0534 to 0.0475. These results link adaptive trust allocation to higher accuracy and more localized updates in long-horizon mathematical reasoning.
The ÌròyìnSpeech Text Corpus: 24,905 Curated Yorùbá Sentences for Speech and Language Technology
Kọ́lá Túbọ̀sún
Anuoluwapo Aremu
Tolúlọpẹ́ Ògúnrẹ̀mí
Iroro Orife
ÌròyìnSpeech is a 42-hour, 80-speaker Yorùbá read-speech corpus whose audio has been distributed by ELRA since 2024. This paper describ… (see more)es the release of its text component: 24,905 unique, hand-verified, tone-marked Yorùbá sentences (275,897 tokens; 15,687 types), curated in 2022 as recording prompts. Roughly 11,000 sentences were adapted from openly licensed news material; the remainder were written in-house to broaden coverage beyond the religious translation that dominates existing Yorùbá corpora. Every sentence was checked by hand for tone-mark accuracy, edited for read-aloud clarity and a neutral register, and localised so that non-Yorùbá personal and place names appear in Yorùbá form. Preparing the text for release surfaced systematic Unicode normalisation failures affecting more than 60% of lines (with precomposed and decomposed forms of the same letter co-occurring within single sentences) which we document and correct. The corpus supports diacritic restoration, grapheme-to-phoneme conversion, TTS front-end development and orthographic research, and serves as a validated prompt set for new recording.
On the Stability and Convergence of Flow Map Training
Ishan Nath
Mathurin Massias
Avishek Joey Bose
Flow maps are a recent method of probabilistic generation that amortize diffusion trajectories into one-step samplers, unlocking efficient g… (see more)eneration. While progress has been made into improving their empirical performance, the understanding of the mechanisms underlying their success remains largely unknown. Specifically, flow maps employ stop gradient operators as a crucial part of their training target, which seem necessary to obtain convergent methods. Adaptive weighting also drastically improves the generation capabilities of flow maps. Although many interpretations have been proposed to support the inclusion of these heuristics, none have analyzed the impact of these assumptions on the training dynamics of flow maps themselves. In this work, we address this issue by providing grounded theory explaining why such tricks are required. Analyzing the functional gradient flow induced by the MeanFlow and Lagrangian objective, we show their convergence to the true flow map, settling the validity of these objectives. For finite updates, we highlight a severe roadblock to the usage of these losses in practice; the possible unboundedness of the update. Adaptive normalization not only provides bounded updates, but combined with a specific weighting, additionally provides convergence guarantees towards a neighborhood of the solution. On synthetic data, we show that our functional gradient flow analysis exactly describes the true training dynamics for flow maps. Experiments on standard image datasets demonstrate that a proper reweighting scheme is essential for strong empirical performance, which is only achieved for a specific range of reweighting strengths.
Clean: Second-order LLM Training at Linear Memory Cost via Nyström Sketching
Beheshteh T. Rakhshan
Sahar Rajabi
Shikai Fang
Sirisha Rambhatla
Training large language models (LLMs) entails a fundamental trade-off: memory-efficient optimizers such as Adam discard cross-parameter curv… (see more)ature, whereas full-curvature methods such as SOAP can accelerate convergence at prohibitive memory costs. We introduce Clean, a memory-efficient and full-curvature optimizer designed to resolve this bottleneck. Clean leverages the randomized Nystrom method to accurately approximate the left and right preconditioners in SOAP, and to reduce the optimizer's memory complexity from quadratic to linear in terms of model dimensions. We subsequently reintegrate the off-subspace components to capture curvature information beyond the low-rank approximation, preserving rich curvature at minimal memory cost. We further propose Q-Clean, a low-precision variant that aggressively compresses optimizer states. Q-Clean reduces optimizer memory consumption by \textbf{over 50\%} compared to Muon when pre-training a LLaMA-1.3B architecture, all while maintaining strong and competitive predictive performance. Notably, Clean operates with a smaller optimizer-state footprint than standard AdamW while reaching AdamW's final performance \textbf{26\% faster} in wall-clock time. Furthermore, our methods uniquely enable the pre-training of a 13B-parameter model on a single 80GB GPU, providing a scalable, efficient, and accessible approach to large-scale model optimization.
Gated Target Propagation for Compositional Generalization in Continual Learning
Continual learning is typically framed as acquiring new knowledge without catastrophically forgetting previous tasks. However, a flexible co… (see more)ntinual learner should also be able to reuse and recombine previously acquired knowledge to rapidly solve novel task compositions. We introduce Gated Target Propagation (GaTaP), a continual learning algorithm in which task-specific gating variables---learned through a closed-form inner loop update---selectively suppress or enhance network modules. Network parameters are learned in a slower timescale outer loop, using the same local difference target propagation error signal as is used for adapting gating variables. We provide tractable experiments on class-incremental learning scenarios for both multilayer perceptron and convolutional network architectures. We show strong performance retention on previously learned tasks, as well as compositional generalization to unseen tasks, achieved through few-shot gain adaptation at inference. We analyze learned gating patterns and find that related tasks exhibit similar gating patterns, suggesting that inferred gates capture meaningful, reusable task structure. Overall, GaTaP provides a powerful framework for jointly ameliorating catastrophic forgetting and enabling few-shot compositional generalization in neural network models.
Gated Transformer World Models for Model-Based Reinforcement Learning
Model-based reinforcement learning has emerged as a promising approach for training sample-efficient policies by learning compact latent dyn… (see more)amics models that simulate, plan, and act through imagination. Recent work has shown that transformer-based world models outperform recurrent architectures due to their ability to capture long-range temporal dependencies. However, standard self-attention mechanisms apply the attention score uniformly across all dimensions of the token's representation, lacking an explicit mechanism to regulate information flow across dimensions. We introduce a gating mechanism for transformer-based world models that implements a feature-wise modulation mechanism within the attention computation. The gating module adaptively scales dimensions in a data-dependent way, enabling the model to selectively control contextual information. Our gating mechanism achieves greater sample-efficient policy learning through normalized mean performance improvements for the Atari 100k benchmark when applied to the STORM and TWISTER transformer world models.
GrayShield: Bit-Level Sanitization for Transformer Model Supply-Chain Security
Armstrong Foundjem
Tsung-Hsien Chuang
Transformer models such as BERT and Vision Transformer~(ViT) achieve strong performance via densely parameterized attention backbones. Howev… (see more)er, the least significant bits~(LSBs) of their 32-bit floating-point weights can be abused as covert channels to conceal malicious payloads, posing a serious threat to the AI model supply chain. We propose \GS (\GSabbr), a lightweight, post-training, zero-data sanitization method that completely replaces the declared mantissa-LSB channel with a Gray-code-guided low-transition sequence. Complete payload-independent overwrite, whether keyed or public, makes the sanitized target bits independent of the embedded payload and gives that declared channel zero capacity. Gray coding supplies overwrite structure, while a keyed per-tensor phase supplies pattern diversity. Benchmarked against seven post-training defenses on four Transformer model presets and two real-world malware payloads, \GSabbr maintains sub-
Path Laplacian Encodings for Directed Graphs
Michael Perlmutter
Bastian Rieck
Directed graphs naturally model many real-world systems in which interactions are asymmetric, such as citation networks, web graphs, and inf… (see more)ormation-flow networks. However, graph learning methods commonly rely on message passing with symmetrized graph representations or positional encodings that only partially exploit edge directionality. We introduce PathLapPE, a novel spectral positional encoding (PE) derived from the path Laplacian on directed graphs. PathLapPE provides node- and edge-level features that encode directional higher-order structure and can be incorporated into standard graph learning architectures. Empirical results on node- and graph-level benchmark tasks show that PathLapPE yields consistent improvements across several architectures, especially when combined with direction-aware message passing. Compared with magnetic Laplacian positional encodings, a widely studied spectral positional encoding for directed graphs, PathLapPE does not require additional fine-tuning of directionality hyperparameters while offering competitive runtime and performance.
Positive interactions and active mentalizing facilitate inter-brain synchronization in couples
Isabella Schneider
Quentin Moreau
Daniel Neudecker
Sabine Christiane Herpertz
High-quality interactions rely on socio-cognitive abilities, which help partners understand and respond to one another. Hyperscanning, which… (see more) measures brain activity in multiple individuals simultaneously, can examine inter-brain synchronization as a possible signature of social coordination. In this study, we explore this synchronization during natural, everyday interactions in romantic couples. Forty-two dyads participated in a naturalistic behavioral paradigm while EEG was recorded. We measured behavioral dyadic reciprocity and negative states, and computed inter-brain synchronization using the circular correlation coefficient (CCorr) across multiple frequency bands (Theta [4-8 Hz], Alpha [8-13 Hz], and Beta [14-31 Hz]). Positive interactions elicited greater dyadic reciprocity and fewer negative states compared to negative interactions. At the neural level, inter-brain synchronization in the beta band was significantly higher during positive compared to negative interactions. Mentalizing compared to counting elicited increased inter-brain synchronization in both alpha and beta bands, indicating that active sociocognitive processing enhances neural alignment between partners. These findings demonstrate that inter-brain dynamics provide a neural signature of both interaction valence and mentalizing engagement during real-world social interactions, putting forward inter-brain synchronization as a sensitive marker of sociocognitive coordination.
PM++: Scaling Differentiable Cosmological Simulators with Dynamic Particle Ownership
Large differentiable particle-mesh simulations support cosmological inference and simulator-coupled learning, but their evolving states can … (see more)exceed single-accelerator memory. Distribution with fixed particle ownership remains memory-limited because deposition halos must cover the maximum cumulative displacement. We present
Steering Speech-Language Models: Training-Free Task Specialization via Contrastive Activation Addition
Yanis Labrak
Sergio Burdisso
Petr Motlicek
Hervé Bredin
Activation steering has proven effective for controlling the behavior of Large Language Models (LLMs) at inference time, but its application… (see more) to SpeechLLMs remains new, and training-free steering approaches for such models are still largely unexplored. We propose a training-free Contrastive Activation Addition (CAA) protocol that derives steering vectors for common speech tasks (e.g. transcription) in SpeechLLMs from a small number of labeled utterances. We showcase that adding these vectors in the representation space, at inference time, enforces better the targeted speech task. We further show that, when combined with prompting, these vectors yield to consistent improvement over prompting alone on most evaluated tasks such as Automatic Speech Recognition (ASR) or Emotion Recognition (ER), and transfer to out-of-domain data. We additionally demonstrate the usefulness of script-normalization directions to enforce the target script of a specific language.
Exploration-Preserving Policy Optimization
Reinforcement learning with verifiable rewards improves reasoning, while the allocation of learning signal shapes which solutions remain acc… (see more)essible under repeated sampling. Group-relative objectives assign equal advantages to equally rewarded responses, making aggregate credit proportional to sampled mode frequency. We introduce Exploration-Preserving Policy Optimization (ExPPO), a lightweight advantage-shaping rule that redistributes credit using prompt-relative, length-normalized response surprisal and prompt pass rate. ExPPO combines bounded shaping with shared normalization to preserve verifier polarity and approximately maintain each prompt group's total absolute sequence-advantage mass. Our analysis characterizes response-level credit allocation alongside sampled mode updates, deriving local conditions for gains in entropy and correct-mode discovery. Experiments show improved in-domain and out-of-domain reasoning coverage, higher aggregate response accuracy, and strong coverage at large sampling budgets. A controlled multi-answer evaluation further demonstrates increased correct-mode yield and gains in diversity among verified-correct responses. Code is available at https://github.com/jinhangzhan/ExPPO