Deploying large language models (LLMs) for structured data extraction in domains such as financial compliance reporting, legal document analytics, and multilingual knowledge base construction is often impractical for smaller teams due to the high cost of running large architectures and the difficulty of preparing large, high-quality datasets. Most recent instruction-tuning studies focus on seven-billion-parameter or larger models, leaving limited evidence on whether much smaller models can work reliably under low-resource, multi-task conditions. This work presents ETLCH, a billion-parameter LLaMA-based model fine-tuned with low-rank adaptation on only a few hundred to one thousand samples per task for JSON extraction, knowledge graph extraction, and named entity recognition. Despite its small scale, ETLCH outperforms strong baselines across most evaluation metrics, with substantial gains observed even at the lowest data scale. These findings demonstrate that well-tuned small models can deliver stable and accurate structured outputs at a fraction of the computational cost, enabling cost-effective and reliable information extraction pipelines in resource-constrained environments.

Modern Large Language Models (LLMs) are excellent at generating synthetic data. However, their performance in sensitive domains such as text detoxification has not received proper attention from the scientific community. This paper explores the possibility of using LLM-generated synthetic toxic data as an alternative to human-generated data for training models for detoxification. Using Llama 3 and Qwen activation-patched models, we generated synthetic toxic counterparts for neutral texts from ParaDetox and SST-2 datasets. Our experiments show that models fine-tuned on synthetic data consistently perform worse than those trained on human data, with a drop in performance of up to 30% in joint metrics. The root cause is identified as a critical lexical diversity gap: LLMs generate toxic content using a small, repetitive vocabulary of insults that fails to capture the nuances and variety of human toxicity. These findings highlight the limitations of current LLMs in this domain and emphasize the continued importance of diverse, human-annotated data for building robust detoxification systems.

Subtrait (latent-trait components) assessment presents a promising path toward enhancing transparency of automated writing scores. We prototype explainability and subtrait scoring with generative language models and show modest correlation between human subtrait and trait scores, and between automated and human subtrait scores. Our approach provides details to demystify scores for educators and students.

Mixture-of-Experts (MoE) has emerged as a promising architecture for modern large language models (LLMs). However, massive parameters impose heavy GPU memory (i.e., VRAM) demands, hindering the widespread adoption of MoE LLMs. Offloading the expert parameters to CPU RAM offers an effective way to alleviate the VRAM requirements for MoE inference. Existing approaches typically cache a small subset of experts in VRAM and dynamically prefetch experts from RAM during inference, leading to significant degradation in inference speed due to the poor cache hit rate and substantial expert loading latency. In this work, we propose MoEpic, an efficient MoE inference system with a novel expert split mechanism. Specifically, each expert is vertically divided into two segments: top and bottom. MoEpic caches the top segment of hot experts, so that more experts will be stored under the limited VRAM budget, thereby improving the cache hit rate. During each layer's inference, MoEpic predicts and prefetches the activated experts for the next layer. Since the top segments of cached experts are exempt from fetching, the loading time is reduced, which allows efficient transfer-computation overlap. Nevertheless, the performance of MoEpic critically depends on the cache configuration (i.e., each layer's VRAM budget and expert split ratio). To this end, we propose a divide-and-conquer algorithm based on fixed-point iteration for adaptive cache configuration. Extensive experiments on popular MoE LLMs demonstrate that MoEpic can save about half of the GPU cost, while lowering the inference latency by about 37.51%-65.73% compared to the baselines.

Existing key-value (KV) cache compression methods typically rely on heuristics, such as uniform cache allocation across layers or static eviction policies, however, they ignore the critical interplays among layer-specific feature patterns and task performance, which can lead to degraded generalization. In this paper, we propose EvolKV, an adaptive framework for layer-wise, task-driven KV cache compression that jointly optimizes the memory efficiency and task performance. By reformulating cache allocation as a multi-objective optimization problem, EvolKV leverages evolutionary search to dynamically configure layer budgets while directly maximizing downstream performance. Extensive experiments on 11 tasks demonstrate that our approach outperforms all baseline methods across a wide range of KV cache budgets on long-context tasks and surpasses heuristic baselines by up to 7 percentage points on GSM8K. Notably, EvolKV achieves superior performance over the full KV cache setting on code completion while utilizing only 1.5% of the original budget, suggesting the untapped potential in learned compression strategies for KV cache budget allocation.

Foundation models are revolutionizing autonomous driving perception, transitioning the field from narrow, task-specific deep learning models to versatile, general-purpose architectures trained on vast, diverse datasets. This survey examines how these models address critical challenges in autonomous perception, including limitations in generalization, scalability, and robustness to distributional shifts. The survey introduces a novel taxonomy structured around four essential capabilities for robust performance in dynamic driving environments: generalized knowledge, spatial understanding, multi-sensor robustness, and temporal reasoning. For each capability, the survey elucidates its significance and comprehensively reviews cutting-edge approaches. Diverging from traditional method-centric surveys, our unique framework prioritizes conceptual design principles, providing a capability-driven guide for model development and clearer insights into foundational aspects. We conclude by discussing key challenges, particularly those associated with the integration of these capabilities into real-time, scalable systems, and broader deployment challenges related to computational demands and ensuring model reliability against issues like hallucinations and out-of-distribution failures. The survey also outlines crucial future research directions to enable the safe and effective deployment of foundation models in autonomous driving systems.

In recent years, VLMs have captured the imagination of the Artificial Intelligence(AI) community, demonstrating an impressive ability to interpret, reason about, and generate content that covers both text and image handling. From answering questions about visual scenes to engaging in multi-modal dialogue, models such as Flamingo [1], PaLI [25], and BLIP-2 [14] are redefining the frontier of vision intelligence. Y et, as these models are widening their application capabilities, a fundamental question emerges: can they truly reason, or are they sophisticated pattern matchers? To explore this question, we turn to the domain of physics--a field that serves as a universal benchmark for logical thoughts of a human being. Physics problems are an ideal testbed for VLMs, as they are multi-modal, combining textual descriptions, mathematical equations, and often clarifying diagrams. A model that can successfully solve these problems must not only understand language and images but also grasp the underlying relationships and principles that govern the physical realm. The challenge, uptil now, has been the lack of accessible tools for this kind of evaluation. Existing benchmarks for scientific reasoning, such as ARC [7] and ScienceQA [17], are often limited to basic text-only question sets, while those that incorporate visual elements, like MathVista [18], frequently depend on complex physics simulators that are computationally expensive for many researchers to deploy, thereby restricting reproducibility.

Distributed and federated learning are essential paradigms for training models across decentralized data sources while preserving privacy, yet communication overhead remains a major bottleneck. This dissertation explores strategies to improve communication efficiency, focusing on model compression, local training, and personalization. We establish a unified framework for biased and unbiased compression operators with convergence guarantees, then propose adaptive local training strategies that incorporate personalization to accelerate convergence and mitigate client drift. In particular, Scafflix balances global and personalized objectives, achieving superior performance under both IID and non-IID settings. We further introduce privacy-preserving pruning frameworks that optimize sparsity while minimizing communication costs, with Cohort-Squeeze leveraging hierarchical aggregation to reduce cross-device overhead. Finally, SymWanda, a symmetric post-training pruning method, enhances robustness under high sparsity and maintains accuracy without retraining. Extensive experiments on benchmarks and large-scale language models demonstrate favorable trade-offs among accuracy, convergence, and communication, offering theoretical and practical insights for scalable, efficient distributed learning.

This paper presents a framework for predicting rare, high-impact outcomes by integrating large language models (LLMs) with a multi-model machine learning (ML) architecture. The approach combines the predictive strength of black-box models with the interpretability required for reliable decision-making. We use LLM-powered feature engineering to extract and synthesize complex signals from unstructured data, which are then processed within a layered ensemble of models including XGBoost, Random Forest, and Linear Regression. The ensemble first produces a continuous estimate of success likelihood, which is then thresholded to produce a binary rare-event prediction. We apply this framework to the domain of Venture Capital (VC), where investors must evaluate startups with limited and noisy early-stage data. The empirical results show strong performance: the model achieves precision between 9.8X and 11.1X the random classifier baseline in three independent test subsets. Feature sensitivity analysis further reveals interpretable success drivers: the startup's category list accounts for 15.6% of predictive influence, followed by the number of founders, while education level and domain expertise contribute smaller yet consistent effects.

The widespread availability of open-source repositories has led to a vast collection of reusable software components, yet their utilization remains manual, error-prone, and disconnected. Developers must navigate documentation, understand APIs, and write integration code, creating significant barriers to efficient software reuse. To address this, we present EnvX, a framework that leverages Agentic AI to agentize GitHub repositories, transforming them into intelligent, autonomous agents capable of natural language interaction and inter-agent collaboration. Unlike existing approaches that treat repositories as static code resources, EnvX reimagines them as active agents through a three-phase process: (1) TODO-guided environment initialization, which sets up the necessary dependencies, data, and validation datasets; (2) human-aligned agentic automation, allowing repository-specific agents to autonomously perform real-world tasks; and (3) Agent-to-Agent (A2A) protocol, enabling multiple agents to collaborate. By combining large language model capabilities with structured tool integration, EnvX automates not just code generation, but the entire process of understanding, initializing, and operationalizing repository functionality. We evaluate EnvX on the GitTaskBench benchmark, using 18 repositories across domains such as image processing, speech recognition, document analysis, and video manipulation. Our results show that EnvX achieves a 74.07% execution completion rate and 51.85% task pass rate, outperforming existing frameworks. Case studies further demonstrate EnvX's ability to enable multi-repository collaboration via the A2A protocol. This work marks a shift from treating repositories as passive code resources to intelligent, interactive agents, fostering greater accessibility and collaboration within the open-source ecosystem.