Publications

Fairer or Just Broken? Differential Privacy in Student Withdrawal Prediction
Carrie Demmans Epp
Nidhi Hegde
Harvey Ngoe Kolle
GlyphBench: A Playground for Language-Model Reinforcement Learning
Roger Creus Castanyer
Matthew James Sargent
Augustine N. Mavor-Parker
We introduce GlyphBench, an environment suite for reinforcement learning (RL) post-training of language-model agents, with over 360 tasks sp… (see more)anning diverse games. GlyphBench renders spatial observations as two-dimensional Unicode grids and connects training, evaluation, and trajectory replay through a unified interface designed to support efficient and reproducible research. We use GlyphBench to study how observation interfaces, reasoning effort, and agent harnesses affect performance, and how RL configurations shape learning dynamics. Our results show that glyph observations outperform native text and pixels in our Craftax experiments, with further gains on several BALROG environments. RL on 100 GlyphBench tasks improves Qwen3.5-4B on held-out Reasoning Gym problems, reaching 63.48% accuracy and outperforming the base model, a math-trained baseline, and a code-trained baseline. These experiments provide empirical evidence that reasoning gains from gameplay can yield stronger transfer than math or code. Together, these results highlight GlyphBench's value as a testbed for systematic research on how language-model agents learn, interact, and generalize.
KV-streams for Efficient Compaction in Agentic Reinforcement Learning
Roger Creus Castanyer
Jonathan Light
Matthew J. Sargent
Augustine N. Mavor-Parker
M. Caccia
Lucas Caccia
Esmeralda S. Whitammer
Minseon Kim
Scaling the horizon of agentic LLMs is bottlenecked by the need to fit ever longer context traces in GPU memory. Context compaction has been… (see more) the most popular mechanism to alleviate this issue, keeping GPU memory constant for a given trace. Unfortunately, most compaction strategies rely on prefilling the LLM context many times over, hindering training throughput. To alleviate this bottleneck and enable efficient trainable compaction, we propose KV-streams, a plug-and-play strategy compatible with any compaction strategy that substantially increases throughput while showing no evidence of hindering performance. KV-streams enable scalable compaction by streaming the KV cache forward rather than flushing it after each compaction. We show that KV-streams enable three different compaction strategies, achieving a 2.6 to 5x wall-clock speedup in training. Beyond efficiency, we find that the streamed KV cache can act as a recurrent state, carrying forward information that has long since disappeared from the context. Specifically, in a controlled setting we show that, contrary to prior work, RL alone is all that is needed for this behavior to emerge. Overall, we show KV-streams to be an efficient and lightweight plug-and-play addition to any post-training pipeline.
KV-streams for Efficient Compaction in Agentic Reinforcement Learning
Roger Creus Castanyer
Jonathan Light
Matthew J. Sargent
Augustine N. Mavor-Parker
M. Caccia
Lucas Caccia
Esmeralda S. Whitammer
Minseon Kim
Scaling the horizon of agentic LLMs is bottlenecked by the need to fit ever longer context traces in GPU memory. Context compaction has been… (see more) the most popular mechanism to alleviate this issue, keeping GPU memory constant for a given trace. Unfortunately, most compaction strategies rely on prefilling the LLM context many times over, hindering training throughput. To alleviate this bottleneck and enable efficient trainable compaction, we propose KV-streams, a plug-and-play strategy compatible with any compaction strategy that substantially increases throughput while showing no evidence of hindering performance. KV-streams enable scalable compaction by streaming the KV cache forward rather than flushing it after each compaction. We show that KV-streams enable three different compaction strategies, achieving a 2.6 to 5x wall-clock speedup in training. Beyond efficiency, we find that the streamed KV cache can act as a recurrent state, carrying forward information that has long since disappeared from the context. Specifically, in a controlled setting we show that, contrary to prior work, RL alone is all that is needed for this behavior to emerge. Overall, we show KV-streams to be an efficient and lightweight plug-and-play addition to any post-training pipeline.
One Rollout Is All You Get: Fully Test-Time Adaptation for GUI Agents
Ziqiang Wang
Zhixiang Chi
Zihuan Jiang
Linqiang Guo
Zhi Liu
GUI agents are deployed with frozen weights and discard everything they experience on the job. Existing ways to update an agent's weights as… (see more)sume something deployment withholds: ground truth, rollouts beyond the single attempt (retries, samples, practice runs), or a learning phase other than deployment. Because GUI actions can be irreversible, a deployed agent gets one attempt per task occurrence, in arrival order, and every attempt counts. No ground truth is available at any point. We define fully test-time adaptation for GUI agents by these constraints and pair it with a minimal weight-space method, SOLO. Auxiliary models read each episode: a judge selects the episodes it deems successful, and a proposer-verifier pair relabels a failed episode's prefix with the subtask that prefix completed. Admitted episodes enter a short sliding window, and each admission updates a small adapter by top-K self-distillation on the agent's own predictions, provided the window holds a judged success. On recurring task streams built from WebArena, VisualWebArena and MobileWorld, SOLO improves on the frozen agent with both UI-TARS-7B and Qwen3-VL-8B, by three to six points of success rate, and exceeds two in-setting memory methods on the web streams.
Shockingly Simple Self-retrospection Improves Agentic Models Without RL
Jonathan Light
Christopher Zhang Cui
Jeonghye Kim
Roger Creus Castanyer
Zhengyan Shi
Xingdi Yuan
Minseon Kim
People learn not only by repeating successful actions, but also by recounting and explaining their experiences, revising their understanding… (see more) to guide future behavior. Can a language-model agent improve its future actions by training only on explanations of its own experience? We investigate this question by studying Retrospection-Only Fine-Tuning (ROFT), a minimal online procedure designed to isolate the effect of explanation-only training on subsequent behavior. The agent attempts a task, observes available feedback, generates a retrospective explanation, and is fine-tuned with a next-token prediction loss on the explanation tokens alone. The procedure uses neither an external teacher nor a reward-based policy update. In software-engineering experiments with Qwen3.5-4B, ROFT is trained on problems with mixed successful and unsuccessful base-model attempts. On held-out SWE-bench Verified and Pro, it reaches 49.2% and 26.8% solve rates after 20 updates without using a verifier, compared with GRPO's 48.0% and 25.3% after 40 updates in the evaluated runs, and makes faster early progress in training time and sampled attempts. It also learns to solve individual tasks on which all 64 sampled base-model attempts failed, showing that learning can begin without any initially successful trajectories. Behavioral analyses find that ROFT indirectly assigns credit to actions, encouraging good actions and discouraging incorrect ones. Moreover, prompting retrospections to emphasize more direct solutions yields shorter subsequent attempts even without an explicit length penalty. Together, these findings show that learning to explain can also improve learning to do, establishing self-generated retrospections as useful training targets and motivating further study of explanation-to-action transfer.
TIPS: A Tiny Inference-time Policy for Scaling in Diffusion Language Models
Masked diffusion language models expose an inference-time choice: which positions become visible after each denoising pass. Many existing de… (see more)coding methods use hand-designed rules or require visibility to grow monotonically. We present a
CollisionGAT: Controller-Agnostic One-Step Collision Screening for Multi-Agent Motion
Before a team of robots moves, each proposed step must be checked for collisions with other robots and with obstacles. We present CollisionG… (see more)AT, a graph-attention network that reads the current and proposed states of moving agents together with locally relevant stationary obstacles and returns one collision-risk score per moving agent. Any controller can use these scores to accept, repair, replan, or postpone a proposed step. We mount CollisionGAT on a continuous path-following controller and on GATeD, an obstacle-blind D* Lite planner that uses typed vetoes to update its planning graphs. Exact geometric checks supply the training labels and independently audit every executed step.
Cool Embeddings: Predicting Galaxy Cluster Cooling Times in IllustrisTNG and TNG-Cluster with AstroCLIP
R. Karam
C. L. Rhea
J. Hlavacek-Larrondo
L. Perreault-Levasseur
M. Prunier
A. Alexandre
H. Choi
Foundation models trained on optical astronomical data have not previously been applied to X-ray cluster thermodynamics. We show that embedd… (see more)ings from AstroCLIP, a contrastive vision-language model trained on optical galaxy images and spectra, retain physically meaningful structure when applied to mock Chandra X-ray images of simulated galaxy clusters - a significant shift in observational modality and physical scale. We target the central cooling time (
From Signals to Trajectories: A Primer on Low-Dimensional Dynamics in Human EEG and MEG
Vanessa Hadid
Annalisa Pascarella
Sara A. Solla
Paul Cisek
Neural activity unfolds not as a set of independent signals, but as a coordinated dynamical process that can be described as a point moving … (see more)through a high-dimensional state space. This perspective has contributed substantially to recent developments in systems neuroscience, especially through studies of directly recorded neuronal population activity, but remains comparatively underused in non-invasive human recordings such as EEG and MEG. In this primer, we present a neural trajectory analysis based on Principal Component Analysis (PCA) as an accessible route into extending the concepts of neural trajectories in state space to human electrophysiology. We explain how EEG/MEG activity can be organized into neural state representations, projected onto axes organized by decreasing covariance, truncated into low-dimensional representations, and visualized as trajectories that capture how distributed activity evolves over time. We show how trajectory geometry can be used to compare conditions and relate neural dynamics to behavior or clinical outcomes. An example of EEG motor execution and imagery, accompanied by a tutorial notebook, illustrates the workflow from conventional EEG summaries to trajectory-based analysis. We also clarify the limits of PCA, including its linear, variance-driven nature, and discuss EEG/MEG-specific challenges such as spatial mixing, preprocessing sensitivity, validation, and overinterpretation. Finally, we situate PCA within a broader family of state-space methods. By making neural trajectories practical and interpretable, this primer offers a guide for bringing the conceptual framework of population-level dynamics into mainstream human cognitive neuroscience.
An evolutionarily conserved interferon-responsive oligodendroglial state emerges during white matter aging
Pieter‐Jan Serneels
Julie D. De Schutter
Luca Masin
A. Collin
Fien Hermans
Julien Cohen‐Adad
Steven Bergmans
Lieve Moons
Abstract White matter dysfunction is a hallmark of brain aging and often precedes overt neurodegeneration, yet the cellular alterations conn… (see more)ecting aging to white matter dysfunction remain poorly understood. Here, we establish the killifish optic nerve as a translational model of white matter aging by integrating transcriptomic, histological and ultrastructural analyses. This tissue recapitulates key molecular and structural features of mammalian white matter aging, including the depletion of oligodendroglial progenitor cells, metabolic dysfunction and reduced myelin thickness. Intriguingly, aging is also associated with the emergence of an interferon-responsive oligodendroglial state. This state is characterized by interferon response and antigen presentation programs that closely resemble those of pathology-associated oligodendroglia described in murine models of central nervous system disease. Our findings identify oligodendroglial dysfunction as a central feature of white matter aging and suggest that this age-associated oligodendroglial state emerges before overt neurodegeneration, providing a potential cellular link between aging and neurodegenerative diseases.
Learning Hierarchical Causal Representations of the Effects of Forcings on Temperature in Climate Models
Machine learning (ML) emulators provide a fast and cost-effective method to simulate climate change scenarios after being trained on Earth S… (see more)ystem Models projections. However, the black-box nature of those data-driven approaches limit the usability and trustworthiness of their outputs and in particular their use as causal attribution tools. Here, we develop a hierarchical causal representation learning framework applied to sea surface temperature fields from a state-of-the-art global climate model. As a key advance over previous work, our framework explicitly models both atmospheric dynamical interactions arising from internal climate variability and forced responses due to changes in atmospheric greenhouse gas and aerosol concentrations. When trained on future climate change scenarios, our method accurately predicts the long-term global mean and regional temperature evolution and shows physically realistic responses to perturbations in greenhouse gas and aerosol concentrations when evaluated on unseen scenarios. Our results underline the potential of causal representation learning frameworks for advancing climate model emulation.