We use cookies to analyze the browsing and usage of our website and to personalize your experience. You can disable these technologies at any time, but this may limit certain functionalities of the site. Read our Privacy Policy for more information.
Setting cookies
You can enable and disable the types of cookies you wish to accept. However certain choices you make could affect the services offered on our sites (e.g. suggestions, personalised ads, etc.).
Essential cookies
These cookies are necessary for the operation of the site and cannot be deactivated. (Still active)
Analytics cookies
Do you accept the use of cookies to measure the audience of our sites?
Multimedia Player
Do you accept the use of cookies to display and allow you to watch the video content hosted by our partners (YouTube, etc.)?
Publications
Revisiting the Goldilocks Zone in Inhomogeneous Networks
We investigate how architectural inhomogeneities—such as biases, layer normalization, and residual connections—affect the curvature of t… (see more)he loss landscape at initialization and its link to trainability. We focus on the Goldilocks zone, a region in parameter space with excess positive curvature, previously associated with improved optimization in homogeneous networks. To extend this analysis, we compare two scaling strategies: weight scaling and softmax temperature scaling.
Our results show that in networks with biases or residual connections, both strategies identify a Goldilocks zone aligned with better training. In contrast, layer normalization leads to lower or negative curvature, yet stable optimization—revealing a disconnect between curvature and trainability. Softmax temperature scaling behaves more consistently across models, making it a more robust probe. Overall, the Goldilocks zone remains relevant in inhomogeneous networks, but its geometry and predictive power depend on architectural choices, particularly normalization.
Tabular foundational models have shown promising in-context learning capabilities on structured data by using training examples as context w… (see more)ithout further parameter adjustments. This emerging approach positions itself as a competitive alternative to traditional gradient-boosted tree methods. However, while biases in conventional machine learning models are well documented, it remains unclear how these biases manifest in Tabular ICL. The paper investigates the fairness implications of Tabular ICL and explores three preprocessing strategies—correlation removal, group-balanced demonstration selection, and uncertainty-based demonstration selection—to address bias. Comprehensive experiments indicate that uncertainty-based demonstration selection consistently enhances group fairness in the predictions. The source code for reproducing the results of this work can be found at https://anonymous.4open.science/r/Fair-TabICL-DD84.
Tabular foundational models have shown promising in-context learning capabilities on structured data by using training examples as context w… (see more)ithout further parameter adjustments. This emerging approach positions itself as a competitive alternative to traditional gradient-boosted tree methods. However, while biases in conventional machine learning models are well documented, it remains unclear how these biases manifest in Tabular ICL. The paper investigates the fairness implications of Tabular ICL and explores three preprocessing strategies—correlation removal, group-balanced demonstration selection, and uncertainty-based demonstration selection—to address bias. Comprehensive experiments indicate that uncertainty-based demonstration selection consistently enhances group fairness in the predictions. The source code for reproducing the results of this work can be found at https://anonymous.4open.science/r/Fair-TabICL-DD84.
The geometry of representations in a neural network can significantly impact downstream generalization. It is unknown how representation geo… (see more)metry changes in large language models (LLMs) over pretraining and post-training. Here, we characterize the evolving geometry of LLM representations using spectral methods (effective rank and eigenspectrum decay). With the OLMo and Pythia model families we uncover a consistent non-monotonic sequence of three distinct geometric phases in pretraining. An initial \warmup phase sees rapid representational compression. This is followed by an "entropy-seeking" phase, characterized by expansion of the representation manifold's effective dimensionality, which correlates with an increase in memorization. Subsequently, a "compression seeking" phase imposes anisotropic consolidation, selectively preserving variance along dominant eigendirections while contracting others, correlating with improved downstream task performance. We link the emergence of these phases to the fundamental interplay of cross-entropy optimization, information bottleneck, and skewed data distribution. Additionally, we find that in post-training the representation geometry is further transformed: Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) correlate with another "entropy-seeking" dynamic to integrate specific instructional or preferential data, reducing out-of-distribution robustness. Conversely, Reinforcement Learning with Verifiable Rewards (RLVR) often exhibits a "compression seeking" dynamic, consolidating reward-aligned behaviors and reducing the entropy in its output distribution. This work establishes the utility of spectral measures of representation geometry for understanding the multiphase learning dynamics within LLMs.
Ultrasound and MRI-based evaluation of relationships between morphological and mechanical properties of the lower lumbar multifidus muscle in chronic low back pain.
We present DoomArena, a security evaluation framework for AI agents. DoomArena is designed on three principles: 1) It is a plug-in framework… (see more) and integrates easily into realistic agentic frameworks like BrowserGym (for web agents) and
Large language model (LLM) agents for web interfaces have advanced rapidly, yet open-source systems still lag behind proprietary agents. Bri… (see more)dging this gap is key to enabling customizable, efficient, and privacy-preserving agents. Two challenges hinder progress: the reproducibility issues in RL and LLM agent training, where results often depend on sensitive factors like seeds and decoding parameters, and the focus of prior work on single-step tasks, overlooking the complexities of web-based, multi-step decision-making.
We address these gaps by providing a statistically driven study of training LLM agents for web tasks. Our two-stage pipeline combines imitation learning from a Llama 3.3 70B teacher with on-policy fine-tuning via Group Relative Policy Optimization (GRPO) on a Llama 3.1 8B student. Through 240 configuration sweeps and rigorous bootstrapping, we chart the first compute allocation curve for open-source LLM web agents. Our findings show that dedicating one-third of compute to teacher traces and the rest to RL improves MiniWoB++ success by 6 points and closes 60% of the gap to GPT-4o on WorkArena, while cutting GPU costs by 45%. We introduce a principled hyperparameter sensitivity analysis, offering actionable guidelines for robust and cost-effective agent training.
The rise of AI agents that can use tools, browse the web and interact with computers on behalf of a user, has sparked strong interest in imp… (see more)roving these capabilities by explicitly fine-tuning the LLMs/VLMs that power these agents. Several researchers have proposed collecting data by letting the agents interact with their environment (e.g., a computer operating system, the web or a collection of APIs exposed as tools), and improve agent performance by fine tuning on this data. In this work, we show that such data collection can be manipulated by adversaries to insert poisoned traces. By modifying just 5% of collected traces, adversaries can embed stealthy bad behaviors into agents—like leaking confidential user information whenever the tool or webpage exposes a trigger. Our results raise important security concerns in the development of AI agents, and underscore the importance of careful scrutiny of all data collection processes used to improve agentic AI.
State entropy regularization has empirically shown better exploration and sample complexity in reinforcement learning (RL). However, its the… (see more)oretical guarantees have not been studied. In this paper, we show that state entropy regularization improves robustness to structured and spatially correlated perturbations. These types of variation are common in transfer learning but often overlooked by standard robust RL methods, which typically focus on small, uncorrelated changes. We provide a comprehensive characterization of these robustness properties, including formal guarantees under reward and transition uncertainty, as well as settings where the method performs poorly. Much of our analysis contrasts state entropy with the widely used policy entropy regularization, highlighting their different benefits. Finally, from a practical standpoint, we illustrate that compared with policy entropy, the robustness advantages of state entropy are more sensitive to the number of rollouts used for policy evaluation.
While large language models (LLMs) have demonstrated remarkable success on a broad range of tasks, math reasoning remains a challenging one.… (see more) One of the approaches for improving math reasoning is self-correction, which designs self-improving loops to let the model correct its own mistakes. However, existing self-correction approaches treat corrections as standalone post-generation refinements, relying on extra prompt and system designs to elicit self-corrections, instead of performing real-time, spontaneous self-corrections in a single pass. To address this, we propose SPOC, a spontaneous self-correction approach that enables LLMs to generate interleaved solutions and verifications in a single inference pass, with generation dynamically terminated based on verification outcomes, thereby effectively scaling inference time compute. SPOC considers a multi-agent perspective by assigning dual roles -- solution proposer and verifier -- to the same model. We adopt a simple yet effective approach to generate synthetic data for fine-tuning, enabling the model to develop capabilities for self-verification and multi-agent collaboration. We further improve its solution proposal and verification accuracy through online reinforcement learning. Experiments on mathematical reasoning benchmarks show that SPOC significantly improves performance. Notably, SPOC boosts the accuracy of Llama-3.1-8B and 70B Instruct models, achieving gains of 8.8% and 11.6% on MATH500, 10.0% and 20.0% on AMC23, and 3.3% and 6.7% on AIME24, respectively.
