We perform a wide variety of experiments and demonstrate that OST outperforms the existing algorithms in the low-shot scenario. On most datasets the method also presents a comparable accuracy with a single training example to the accuracy obtained by the other methods for the entire set of domainA images.

This technical report presents a comprehensive formal verification approach for probabilistic agent systems modeling ballistic rocket flight trajectories using Probabilistic Alternating-Time Temporal Logic (PATL). We describe an innovative verification framework specifically designed for analyzing critical safety properties of ballistic rockets engineered to achieve microgravity conditions for scientific experimentation. Our model integrates authentic flight telemetry data encompassing velocity vectors, pitch angles, attitude parameters, and GPS coordinates to construct probabilistic state transition systems that rigorously account for environmental stochasticity, particularly meteorological variability. We formalize mission-critical safety properties through PATL specifications to systematically identify trajectory deviation states where the rocket risks landing in prohibited or hazardous zones. The verification framework facilitates real-time safety monitoring and enables automated intervention mechanisms, including emergency engine disengagement protocols, when predefined safety thresholds are exceeded. Experimental validation demonstrates the practical effectiveness and reliability of our approach in ensuring mission safety while maintaining scientific mission objectives.

Reinforcement Learning (RL) can mitigate the causal confusion and distribution shift inherent to imitation learning (IL). However, applying RL to end-to-end autonomous driving (E2E-AD) remains an open problem for its training difficulty, and IL is still the mainstream paradigm in both academia and industry. Recently Model-based Reinforcement Learning (MBRL) have demonstrated promising results in neural planning; however, these methods typically require privileged information as input rather than raw sensor data. We fill this gap by designing Raw2Drive, a dual-stream MBRL approach. Initially, we efficiently train an auxiliary privileged world model paired with a neural planner that uses privileged information as input. Subsequently, we introduce a raw sensor world model trained via our proposed Guidance Mechanism, which ensures consistency between the raw sensor world model and the privileged world model during rollouts. Finally, the raw sensor world model combines the prior knowledge embedded in the heads of the privileged world model to effectively guide the training of the raw sensor policy. Raw2Drive is so far the only RL based end-to-end method on CARLA Leaderboard 2.0, and Bench2Drive and it achieves state-of-the-art performance.

Automated plankton recognition models face significant challenges during real-world deployment due to distribution shifts (Out-of-Distribution, OoD) between training and test data. This stems from plankton's complex morphologies, vast species diversity, and the continuous discovery of novel species, which leads to unpredictable errors during inference. Despite rapid advancements in OoD detection methods in recent years, the field of plankton recognition still lacks a systematic integration of the latest computer vision developments and a unified benchmark for large-scale evaluation. T o address this, this paper meticulously designed a series of OoD benchmarks simulating various distribution shift scenarios based on the DYB-PlanktonNet dataset [27], and systematically evaluated twenty-two OoD detection methods. Extensive experimental results demonstrate that the ViM [57] method significantly outperforms other approaches in our constructed benchmarks, particularly excelling in Far-OoD scenarios with substantial improvements in key metrics. This comprehensive evaluation not only provides a reliable reference for algorithm selection in automated plankton recognition but also lays a solid foundation for future research in plankton OoD detection. T o our knowledge, this study marks the first large-scale, systematic evaluation and analysis of Out-of-Distribution data detection methods in plankton recognition. Code is available at https://github.com/BlackJack0083/

Academic programs are increasingly recognizing the importance of personal and professional skills and their critical role alongside technical expertise in preparing students for future success in diverse career paths. With this growing demand comes the need for scalable systems to measure, evaluate, and develop these skills. Situational Judgment Tests (SJTs) offer one potential avenue for measuring these skills in a standardized and reliable way, but open-response SJTs have traditionally relied on trained human raters for evaluation, presenting operational challenges to delivering SJTs at scale. Past attempts at developing NLP-based scoring systems for SJTs have fallen short due to issues with construct validity of these systems. In this article, we explore a novel approach to extracting construct-relevant features from SJT responses using large language models (LLMs). We use the Casper SJT to demonstrate the efficacy of this approach. This study sets the foundation for future developments in automated scoring for personal and professional skills.

We demonstrate how AI agents can coordinate to deceive oversight systems using automated interpretability of neural networks. Using sparse autoencoders (SAEs) as our experimental framework, we show that language models (Llama, DeepSeek R1, and Claude 3.7 Sonnet) can generate deceptive explanations that evade detection. Our agents employ steganographic methods to hide information in seemingly innocent explanations, successfully fooling oversight models while achieving explanation quality comparable to reference labels. We further find that models can scheme to develop deceptive strategies when they believe the detection of harmful features might lead to negative consequences for themselves. All tested LLM agents were capable of deceiving the overseer while achieving high interpretability scores comparable to those of reference labels. We conclude by proposing mitigation strategies, emphasizing the critical need for robust understanding and defenses against deception.

The subjective evaluation of early stage engineering designs, such as conceptual sketches, traditionally relies on human experts. However, expert evaluations are time-consuming, expensive, and sometimes inconsistent. Recent advances in vision-language models (VLMs) offer the potential to automate design assessments, but it is crucial to ensure that these AI ``judges'' perform on par with human experts. However, no existing framework assesses expert equivalence. This paper introduces a rigorous statistical framework to determine whether an AI judge's ratings match those of human experts. We apply this framework in a case study evaluating four VLM-based judges on key design metrics (uniqueness, creativity, usefulness, and drawing quality). These AI judges employ various in-context learning (ICL) techniques, including uni- vs. multimodal prompts and inference-time reasoning. The same statistical framework is used to assess three trained novices for expert-equivalence. Results show that the top-performing AI judge, using text- and image-based ICL with reasoning, achieves expert-level agreement for uniqueness and drawing quality and outperforms or matches trained novices across all metrics. In 6/6 runs for both uniqueness and creativity, and 5/6 runs for both drawing quality and usefulness, its agreement with experts meets or exceeds that of the majority of trained novices. These findings suggest that reasoning-supported VLM models can achieve human-expert equivalence in design evaluation. This has implications for scaling design evaluation in education and practice, and provides a general statistical framework for validating AI judges in other domains requiring subjective content evaluation.

Riddle-solving requires advanced reasoning skills, pushing LLMs to engage in abstract thinking and creative problem-solving, often revealing limitations in their cognitive abilities. In this paper, we examine the riddle-solving capabilities of LLMs using a multiple-choice format, exploring how different prompting techniques impact performance on riddles that demand diverse reasoning skills. To enhance results, we introduce RISCORE (RIddle Solving with COntext REcontruciton) a novel fully automated prompting method that generates and utilizes contextually reconstructed sentence-based puzzles in conjunction with the original examples to create few-shot exemplars. Our experiments demonstrate that RISCORE significantly improves the performance of language models in both vertical and lateral thinking tasks, surpassing traditional exemplar selection strategies across a variety of few-shot settings.

In industrial contexts, effective workforce allocation is crucial for operational efficiency. This paper presents an ongoing project focused on developing a decision-making tool designed for workforce allocation, emphasizing the explainability to enhance its trustworthiness. Our objective is to create a system that not only optimises the allocation of teams to scheduled tasks but also provides clear, understandable explanations for its decisions, particularly in cases where the problem is infeasible. By incorporating human-in-the-loop mechanisms, the tool aims to enhance user trust and facilitate interactive conflict resolution. We implemented our approach on a prototype tool/digital demonstrator intended to be evaluated on a real industrial scenario both in terms of performance and user acceptability.

Today's largest foundation models have increasingly general capabilities, yet when used as agents, they often struggle with simple reasoning and decision-making tasks, even though they possess good factual knowledge of the task and how to solve it. In this paper, we present a benchmark to pressure-test these models' multimodal decision-making capabilities in the very long-context regime (up to one million tokens) and investigate whether they can learn from a large number of expert demonstrations in their context. We evaluate a wide range of state-of-the-art frontier models as policies across a battery of simple interactive decision-making tasks: playing tic-tac-toe, chess, and Atari, navigating grid worlds, solving crosswords, and controlling a simulated cheetah. We measure the performance of Claude 3.5 Sonnet, Gemini 1.5 Flash, Gemini 1.5 Pro, GPT-4o, o1-mini, and o1-preview under increasing amounts of expert demonstrations in the context $\unicode{x2013}$ from no demonstrations up to 512 full episodes, pushing these models' multimodal long-context reasoning capabilities to their limits. Across our tasks, today's frontier models rarely manage to fully reach expert performance, showcasing the difficulty of our benchmark. Presenting more demonstrations often has little effect, but some models steadily improve with more demonstrations on a few tasks. We investigate the effect of encoding observations as text or images and the impact of chain-of-thought prompting. Overall, our results suggest that even today's most capable models often struggle to imitate desired behavior by generalizing purely from in-context demonstrations. To help quantify the impact of other approaches and future innovations aiming to tackle this problem, we open source our benchmark that covers the zero-, few-, and many-shot regimes in a unified evaluation.