Clinical conversations mix explicit directives (order a chest X-ray) with implicit reasoning (the cough worsened overnight, we should check for pneumonia). Many systems rely on LLM rewriting, adding latency, instability, and opacity that hinder real-time ordering. We present JEDA (Joint Embedding for Direct and Ambient clinical orders), a domain-initialized bi-encoder that retrieves canonical orders directly and, in a query-free mode, encodes a short rolling window of ambient dialogue to trigger retrieval. Initialized from PubMedBERT and fine-tuned with a duplicate-safe contrastive objective, JEDA aligns heterogeneous expressions of intent to shared order concepts. Training uses constrained LLM guidance to tie each signed order to complementary formulations (command only, context only, command+context, context+reasoning), producing clearer inter-order separation, tighter query extendash order coupling, and stronger generalization. The query-free mode is noise-resilient, reducing sensitivity to disfluencies and ASR errors by conditioning on a short window rather than a single utterance. Deployed in practice, JEDA yields large gains and substantially outperforms its base encoder and recent open embedders (Linq Embed Mistral, SFR Embedding, GTE Qwen, BGE large, Embedding Gemma). The result is a fast, interpretable, LLM-free retrieval layer that links ambient context to actionable clinical orders in real time.

We introduce GateSkip, a simple residual-stream gating mechanism that enables token-wise layer skipping in decoder-only LMs. Each Attention/MLP branch is equipped with a sigmoid-linear gate that condenses the branch's output before it re-enters the residual stream. During inference we rank tokens by the gate values and skip low-importance ones using a per-layer budget. While early-exit or router-based Mixture-of-Depths models are known to be unstable and need extensive retraining, our smooth, differentiable gates fine-tune stably on top of pretrained models. On long-form reasoning, we save up to 15% compute while retaining over 90% of baseline accuracy. For increasingly larger models, this tradeoff improves drastically. On instruction-tuned models we see accuracy gains at full compute and match baseline quality near 50% savings. The learned gates give insight into transformer information flow (e.g., BOS tokens act as anchors), and the method combines easily with quantization, pruning, and self-speculative decoding.

Prompt injection attacks deceive a large language model into completing an attacker-specified task instead of its intended task by contaminating its input data with an injected prompt, which consists of injected instruction(s) and data. Localizing the injected prompt within contaminated data is crucial for post-attack forensic analysis and data recovery. Despite its growing importance, prompt injection localization remains largely unexplored. In this work, we bridge this gap by proposing PromptLocate, the first method for localizing injected prompts. PromptLocate comprises three steps: (1) splitting the contaminated data into semantically coherent segments, (2) identifying segments contaminated by injected instructions, and (3) pinpointing segments contaminated by injected data. We show PromptLocate accurately localizes injected prompts across eight existing and eight adaptive attacks.

Monte Carlo (TSMC) to provide scalable stepwise supervision using small LMs. This yields more efficient training, improved stability, and higher accuracy. OpenAI's o1 series (OpenAI, 2024) introduce inference-time scaling by increasing the length of the Chain-of-Thought (CoT) (Wei et al., 2022) reasoning process. Despite their empirical success, RL approaches that generate the entire reasoning chain in a single forward pass face notable limitations, including CoT derailment, where the reasoning trajectory drifts off course due to accumulated errors, and the inherent challenges of long-horizon RL with sparse outcome rewards. This sequential scaling strategy, i.e., simply extending the CoT length, can therefore be insufficient (Y ang et al., 2025). To improve planning quality, we introduce Multi-Path Plan Aggregation (MPP A). For each planning step, the model generates multiple alternative plans and aggregates them into an improved plan before proceeding to the subsequent execution steps. Beyond enhancing planning, we identify a fundamental challenge in credit assignment for long-horizon policy learning (Kaelbling et al., 1996). Existing RL fine-tuning frameworks struggle to provide effective process-level supervision (Guo et al., 2025). First, evaluating the correctness of intermediate steps is inherently difficult. Automated annotation using LLM judges (Gu et al., 2024) often yield unreliable or noisy signals Second, introducing a separate process reward model (PRM) adds complexity. We then define the process preference between two candidate continuations at the same step by comparing their incremental log-weights. We repurpose Twisted Sequential Monte Carlo (TSMC) to provide process-level preferences for online Step-DPO training. Results show that our approach consistently outperforms both distillation-based long-CoT methods and RL methods that rely solely on outcome rewards. The Chain-of-Thought trajectories can be lengthy and the positions of the first error vary considerably, making outcome-based RL fine-tuning inefficient. Training long trajectories with outcome rewards is highly inefficient.

The role of world knowledge has been particularly crucial to predict the discourse connective that marks the discourse relation between two arguments, with language models (LMs) being generally successful at this task. We flip this premise in our work, and instead study the inverse problem of understanding whether discourse connectives can inform LMs about the world. To this end, we present WUGNECTIVES, a dataset of 8,880 stimuli that evaluates LMs' inferences about novel entities in contexts where connectives link the entities to particular attributes. On investigating 17 different LMs at various scales, and training regimens, we found that tuning an LM to show reasoning behavior yields noteworthy improvements on most connectives. At the same time, there was a large variation in LMs' overall performance across connective type, with all models systematically struggling on connectives that express a concessive meaning. Our findings pave the way for more nuanced investigations into the functional role of language cues as captured by LMs. We release WUGNECTIVES at https://github.com/sheffwb/wugnectives.

Large Language Models (LLMs) are powerful reasoners in natural language, but their actions are typically confined to outputting vocabulary tokens. As a result, interactions with external environments -- such as symbolic operators or simulators -- must be expressed through text in predefined formats, parsed, and routed to external interfaces. This overloads the model's language with both reasoning and control duties, and requires a hand-crafted parser, external to the LLM. To address this, we decouple environment interactions from language by internalizing them in an Expanded Action space (ExpA), beyond the vocabulary. The model starts reasoning in the default language environment, but may trigger routing actions and switch to an external environment at any time. From there, the model can only invoke environment-specific actions, receive feedback from the environment, and potentially route back to language as a result. To promote effective exploration of the expanded action space and new environments, we introduce ExpA Reinforcement Learning (EARL) with counterfactual policy optimization. On tasks requiring multi-turn interactions and contingent planning, EARL outperforms strong baselines with vocabulary-constrained actions. It performs robustly across calculator-based multi-task learning and, in the partially observed sorting problem, achieves perfect Sort-4 accuracy while self-discovering an efficient algorithm competitive with classical designs.

Post-training for reasoning of large language models (LLMs) increasingly relies on verifiable rewards: deterministic checkers that provide 0-1 correctness signals. While reliable, such binary feedback is brittle--many tasks admit partially correct or alternative answers that verifiers under-credit, and the resulting all-or-nothing supervision limits learning. Reward models offer richer, continuous feedback, which can serve as a complementary supervisory signal to verifiers. We introduce HERO (Hybrid Ensemble Reward Optimization), a reinforcement learning framework that integrates verifier signals with reward-model scores in a structured way. HERO employs stratified normalization to bound reward-model scores within verifier-defined groups, preserving correctness while refining quality distinctions, and variance-aware weighting to emphasize challenging prompts where dense signals matter most. Across diverse mathematical reasoning benchmarks, HERO consistently outperforms RM-only and verifier-only baselines, with strong gains on both verifiable and hard-to-verify tasks. Our results show that hybrid reward design retains the stability of verifiers while leveraging the nuance of reward models to advance reasoning.

This manuscript has been accepted in IEEE Transactions on Industrial Informatics. Personal use of this material is permitted. Abstract--Accurate battery capacity estimation is key to alleviating consumer concerns about battery performance and reliability of electric vehicles (EVs). However, practical data limitations imposed by stringent privacy regulations and labeled data shortages hamper the development of generalizable capacity estimation models that remain robust to real-world data distribution shifts. While self-supervised learning can leverage unlabeled data, existing techniques are not particularly designed to learn effectively from challenging field data--let alone from privacy-friendly data, which are often less feature-rich and noisier . In this work, we propose a first-of-its-kind capacity estimation model based on self-supervised pre-training, developed on a large-scale dataset of privacy-friendly charging data snippets from real-world EV operations. Our pre-training framework, snippet similarity-weighted masked input reconstruction, is designed to learn rich, generalizable representations even from less feature-rich and fragmented privacy-friendly data. Our key innovation lies in harnessing contrastive learning to first capture high-level similarities among fragmented snippets that otherwise lack meaningful context. With our snippet-wise contrastive learning and subsequent similarity-weighted masked reconstruction, we are able to learn rich representations of both granular charging patterns within individual snippets and high-level associative relationships across different snippets. Bolstered by this rich representation learning, our model consistently outperforms state-of-the-art baselines, achieving 31.9%

Rock capturing with standard excavator buckets is a challenging task typically requiring the expertise of skilled operators. Unlike soil digging, it involves manipulating large, irregular rocks in unstructured environments where complex contact interactions with granular material make model-based control impractical. Existing autonomous excavation methods focus mainly on continuous media or rely on specialized grippers, limiting their applicability to real-world construction sites. This paper introduces a fully data-driven control framework for rock capturing that eliminates the need for explicit modeling of rock or soil properties. Robustness is enhanced through extensive domain randomization of rock geometry, density, and mass, as well as the initial configurations of the bucket, rock, and goal position. To the best of our knowledge, this is the first study to develop and evaluate an RL-based controller for the rock capturing task. Experimental results show that the policy generalizes well to unseen rocks and varying soil conditions, achieving high success rates comparable to those of human participants while maintaining machine stability. Corresponding author Email address: amirmasoud.molaei@tuni.fi Keywords: Excavators, Automatic rock capturing, Reinforcement learning, High-fidelity simulation, Guiding Reward Formulation, Non-prehensile manipulation 1. Introduction Autonomous excavation holds a great promise in addressing increasing demands of the mining and construction industries, two of the largest and most essential sectors worldwide. The excavator is one of the most widely used and versatile heavy-duty mobile machines (HDMMs), which is typically operated through a hydraulic system. Excavators are utilized for a wide range of earth-moving tasks, including digging, trenching, grading, and in particular material handling. Despite their versatility, traditional manual operation of excavators can result in low efficiency, increased physical strain on operators, and exposure to hazardous environments like open-pit mines. These challenges underscore the need for automation to enhance safety and productivity. An excavator is primarily composed of three major components, the traveling body, swing body, and the front digging manipulator. The digging manipulator, includes three main parts, boom, arm, and bucket, which are actuated by hydraulic cylinders. Additionally, joints connect the swing body, boom, arm, and bucket, allowing for flexible and precise motion [1, 2, 3, 4].

Survival analysis is central in many applications across medicine, engineering, economics and finance. It concerns time-to-event modeling: given a process that can generate an event of interest (e.g., death from disease, failure due to wear), the goal is to estimate the probability that an event occurs at any time t > 0 for an individual described by some input variables (or features, or covariates). Unlike standard regression settings, survival data are characterized by censoring, which means that for some instances, the exact event time is not observed (for example, when individuals remain event-free at the end of the study), and only the last recorded follow-up time is available. Traditional approaches to survival modeling rely on strong statistical assumptions linking input variables and risk. The Cox proportional hazards (CoxPH) model [1] remains the most widely used and best established method. The proportional hazards assumption implies that the instantaneous risk of event for two individuals differs by a constant factor over time. The CoxPH model is also linear, making it clear how each single input variable affects the outcome, but at the expense of missing interactions between features. In its original form, this relation is modeled through a linear regression on the features, though many extensions have been developed to relax linearity and improve performance in high-dimensional settings [2-4]. Despite its success, Cox regression is limited by the proportional hazards (PH) assumption, which is often unrealistic.