Large Language Models (LLMs) with vast context windows offer new avenues for in-context learning (ICL), where providing many examples ("many-shot" prompting) is often assumed to enhance performance. We investigate this assumption for the complex task of code translation. Through a large-scale empirical study of over 90,000 translations, we systematically evaluate the impact of scaling in-context examples from zero-shot to many-shot configurations of up to 625 examples, with prompts spanning from approximately 100,000 to 800,000 tokens. Our findings reveal a "many-shot paradox": while static similarity metrics may modestly improve with more examples, functional correctness consistently peaks with few-shot prompting (5-25 examples). Providing substantially more examples often degrades this crucial functional performance. This study highlights that for code translation, the quality of a few well-chosen examples outweighs sheer quantity, challenging the universal efficacy of "more is better" for ICL and underscoring the task-dependent nature of optimal prompting strategies. Our results have significant implications for effectively leveraging LLMs in software engineering.

Despite the renewed interest in zero-shot visual classification due to the rise of Multimodal Large Language Models (MLLMs), the problem of evaluating free-form responses of auto-regressive models remains a persistent challenge. Most existing works focus on language-only tasks or don't consider Multiple Choice Questions (MCQs) beyond 5-way options, both of which are critical capabilities to solve tasks in Fine-Grained Visual Classification (FGVC) where choice counts are in the hundreds to thousands and the choices are highly related. Furthermore, in this highly multi-way MCQ setting it is not clear how to extend LLM choice extraction to retrieval-based problems, where computing probabilities over the choice set is computationally costly. In this work we investigate nlg2choice, a simple two-stage method which first asks the MLLM an open-ended question for the task with minimal constraints, then uses text-only constrained decoding to predict the most likely choice. In retrieval settings, we compute the probability of the constrained response taking that choice with an early stopping method to significantly improve throughput. Our results show improvement over a suite of seven fine-grained visual datasets when evaluating in terms of classification and retrieval, and show that this performance holds over the various ways that users of LLMs can implement tasks in natural language.

As the markets for unmanned aerial vehicles (UAVs) and mixed reality (MR) headsets continue to grow, recent research has increasingly explored their integration, which enables more intuitive, immersive, and situationally aware control systems. We present IGUANA, an MR-based immersive guidance, navigation, and control system for consumer UAVs. IGUANA introduces three key elements beyond conventional control interfaces: (1) a 3D terrain map interface with draggable waypoint markers and live camera preview for high-level control, (2) a novel spatial control metaphor that uses a virtual ball as a physical analogy for low-level control, and (3) a spatial overlay that helps track the UAV when it is not visible with the naked eye or visual line of sight is interrupted. We conducted a user study to evaluate our design, both quantitatively and qualitatively, and found that (1) the 3D map interface is intuitive and easy to use, relieving users from manual control and suggesting improved accuracy and consistency with lower perceived workload relative to conventional dual-stick controller, (2) the virtual ball interface is intuitive but limited by the lack of physical feedback, and (3) the spatial overlay is very useful in enhancing the users' situational awareness.

Abstract--Feature selection eliminates redundancy among features to improve downstream task performance while reducing computational overhead. Existing methods often struggle to capture intricate feature interactions and adapt across diverse application scenarios. Recent advances employ generative intelligence to alleviate these drawbacks. However, these methods remain constrained by permutation sensitivity in embedding and reliance on convexity assumptions in gradient-based search. T o address these limitations, our initial work introduces a novel framework that integrates permutation-invariant embedding with policy-guided search. Although effective, it still left opportunities to adapt to realistic distributed scenarios. In practice, data across local clients is highly imbalanced, heterogeneous and constrained by strict privacy regulations, limiting direct sharing. These challenges highlight the need for a framework that can integrate feature selection knowledge across clients without exposing sensitive information. In this extended journal version, we advance the framework from two perspectives: 1) developing a privacy-preserving knowledge fusion strategy to derive a unified representation space without sharing sensitive raw data. The results further demonstrate its strong generalization ability in federated learning scenarios. The code and data are publicly available https://anonymous.4open.science/r/FedCAPS-08BF. Index T erms--Automated Feature Selection; Representation Learning; Reinforcement Learning, Federated Learning. EA TURE selection removes redundant and irrelevant features to improve both predictive performance and computational efficiency in downstream tasks. Despite the growing dominance of deep learning, feature selection remains indispensable in scenarios characterized by high-dimensional data, the need for interpretability, and limited resource constraints.

Reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) are important techniques to align large language models (LLM) with human preference. However, the quality of RLHF and DPO training is seriously compromised by \textit{\textbf{C}orrupted} preference, reward \textit{\textbf{O}veroptimization}, and bias towards \textit{\textbf{V}erbosity}. To our knowledge, most existing works tackle only one of these important issues, and the few other works require much computation to estimate multiple reward models and lack theoretical guarantee of generalization ability. In this work, we propose RLHF-\textbf{COV} and DPO-\textbf{COV} algorithms that can simultaneously mitigate these three issues, in both offline and online settings. This ability is theoretically demonstrated by obtaining length-regularized generalization error rates for our DPO-COV algorithms trained on corrupted data, which match the best-known rates for simpler cases with clean data and without length regularization. Moreover, our DPO-COV algorithm is simple to implement without reward estimation, and is proved to be equivalent to our RLHF-COV algorithm, which directly implies the equivalence between the vanilla RLHF and DPO algorithms. Experiments demonstrate the effectiveness of our DPO-COV algorithms under both offline and online settings.

Large-scale public deliberations generate thousands of free-form contributions that must be synthesized into representative and neutral summaries for policy use. While LLMs have been shown as a promising tool to generate summaries for large-scale deliberations, they also risk underrepresenting minority perspectives and exhibiting bias with respect to the input order, raising fairness concerns in high-stakes contexts. Studying and fixing these issues requires a comprehensive evaluation at a large scale, yet current practice often relies on LLMs as judges, which show weak alignment with human judgments. To address this, we present DeliberationBank, a large-scale human-grounded dataset with (1) opinion data spanning ten deliberation questions created by 3,000 participants and (2) summary judgment data annotated by 4,500 participants across four dimensions (representativeness, informativeness, neutrality, policy approval). Using these datasets, we train DeliberationJudge, a fine-tuned DeBERTa model that can rate deliberation summaries from individual perspectives. DeliberationJudge is more efficient and more aligned with human judgements compared to a wide range of LLM judges. With DeliberationJudge, we evaluate 18 LLMs and reveal persistent weaknesses in deliberation summarization, especially underrepresentation of minority positions. Our framework provides a scalable and reliable way to evaluate deliberation summarization, helping ensure AI systems are more representative and equitable for policymaking.

Converting natural language questions into SQL queries enables non-expert users to interact with relational databases and has long been a central task for natural language interfaces to data. While the WikiSQL dataset played a key role in early text-to-SQL research, its usage has declined due to structural and annotation issues, including case sensitivity inconsistencies, data type mismatches, syntax errors, and unanswered questions. We present LLMSQL, a systematic revision and transformation of WikiSQL designed for the large language model era. We classify these errors and implement automated methods for cleaning and re-annotation. To assess the impact of these improvements, we evaluated multiple large language models, including Gemma 3, LLaMA 3.2, Mistral 7B, gpt-oss 20B, Phi-3.5 Mini, Qwen 2.5, OpenAI o4-mini, DeepSeek-R1, and others. Notably, DeepSeek-R1 achieves 88.40% accuracy in a zero-shot setting, and models under 10B parameters surpass 90% accuracy after fine-tuning. Rather than serving as an update, LLMSQL is introduced as an LLM-ready benchmark. Unlike the original WikiSQL, which was tailored for pointer-network models selecting tokens from input, LLMSQL provides clean natural language questions and full SQL queries as plain text, enabling straightforward generation and evaluation for modern natural-language-to-SQL models.

Large Language Models (LLMs) often fail on multiple-choice questions (MCQs) despite demonstrating correct knowledge in other contexts, such as free-form generation. To investigate the mechanism underlying this knowledge-prediction gap on MCQs and alleviate it, we conduct a probing analysis and find that residual streams in certain layers contain a subspace spanned by two important bases: a \emph{knowledge basis} that encodes the probability of the ground-truth answer for a given MCQ and a \emph{prediction basis} that encodes the probability of the answer choice predicted by the model. We observe that incorrect predictions arise from a misalignment of the model's hidden states along these two bases. Hence, we introduce \textbf{KAPPA} (Knowledge-Aligned Prediction through Projection-based Adjustment), a parameter-free intervention that transforms the hidden states to align the prediction coordinate with the knowledge coordinate within this subspace. Experiments on binary-choice reformulations of Big-Bench-Hard and ARC-Challenge show that KAPPA substantially improves accuracy and consistently outperforms baselines. While optimal subspaces differ across tasks, subspaces generalize to some extent, as supported by cross-dataset experiments. Moreover, KAPPA extends its effectiveness to free-form questions beyond MCQs. Our work provides a new geometric understanding of the knowledge-prediction gap and offers a practical method for better aligning model behavior with its latent knowledge.

We propose a reduced-space formulation for optimizing over trained neural networks where the network's outputs and derivatives are evaluated on a GPU. To do this, we treat the neural network as a "gray box" where intermediate variables and constraints are not exposed to the optimization solver. Compared to the full-space formulation, in which intermediate variables and constraints are exposed to the optimization solver, the reduced-space formulation leads to faster solves and fewer iterations in an interior point method. We demonstrate the benefits of this method on two optimization problems: Adversarial generation for a classifier trained on MNIST images and security-constrained optimal power flow with transient feasibility enforced using a neural network surrogate.

Large vision models differ widely in architecture and training paradigm, yet we lack principled methods to determine which aspects of their representations are shared across families and which reflect distinctive computational strategies. We leverage a suite of representational similarity metrics, each capturing a different facet-geometry, unit tuning, or linear decodability-and assess family separability using multiple complementary measures. Metrics preserving geometry or tuning (e.g., RSA, Soft Matching) yield strong family discrimination, whereas flexible mappings such as Linear Predictivity show weaker separation. These findings indicate that geometry and tuning carry family-specific signatures, while linearly decodable information is more broadly shared. To integrate these complementary facets, we adapt Similarity Network Fusion (SNF), a method inspired by multi-omics integration. SNF achieves substantially sharper family separation than any individual metric and produces robust composite signatures. Clustering of the fused similarity matrix recovers both expected and surprising patterns: supervised ResNets and ViTs form distinct clusters, yet all self-supervised models group together across architectural boundaries. Hybrid architectures (ConvNeXt, Swin) cluster with masked autoencoders, suggesting convergence between architectural modernization and reconstruction-based training. This biology-inspired framework provides a principled typology of vision models, showing that emergent computational strategies-shaped jointly by architecture and training objective-define representational structure beyond surface design categories.