While 4-bit quantization has emerged as a memory-optimal choice for non-reasoning models and zero-shot tasks across scales, we show that this universal prescription fails for reasoning models, where the KV cache rather than model size can dominate memory. Through systematic experiments across 1,700 inference scenarios on AIME25 and GPQA-Diamond, we find a scale-dependent trade-off: models with an effective size below 8-bit 4B parameters achieve better accuracy by allocating memory to more weights rather than longer generation, while larger models achieve better accuracy by allocating memory to longer generations. This scale threshold also determines when parallel scaling becomes memory-efficient and whether KV cache eviction outperforms KV quantization. Our findings show that memory optimization for LLMs cannot be scale-agnostic, while providing principled guidelines: for small reasoning models, prioritize model capacity over test-time compute, while for larger ones, maximize test-time compute. Our results suggest that optimizing reasoning models for deployment requires fundamentally different strategies from those established for non-reasoning models.Figure 1: Memory vs. Accuracy for serial test-time scaling on AIME25. The plot illustrates the trade-off between pass@1 accuracy and total memory (weights + KV cache) for the Qwen3 family. Model weights are quantized to 4-and 8-bit using GPTQ. Along each curve, the KV cache grows as the generation length increases via budget forcing. For models effectively smaller than an 8-bit 4B, increasing the token budget to saturation is memory-inefficient. Furthermore, for mathematical reasoning, higher weight precision (8-and 16-bit) proves more memory-efficient than 4-bit. This work was done during an internship at KRAFTON.

Translating C code into safe Rust is an effective way to ensure its memory safety. Compared to rule-based translation which produces Rust code that remains largely unsafe, LLM-based methods can generate more idiomatic and safer Rust code because LLMs have been trained on vast amount of human-written idiomatic code. Although promising, existing LLM-based methods still struggle with project-level C-to-Rust translation. They typically partition a C project into smaller units (\eg{} functions) based on call graphs and translate them bottom-up to resolve program dependencies. However, this bottom-up, unit-by-unit paradigm often fails to translate pointers due to the lack of a global perspective on their usage. To address this problem, we propose a novel C-Rust Pointer Knowledge Graph (KG) that enriches a code-dependency graph with two types of pointer semantics: (i) pointer-usage information which record global behaviors such as points-to flows and map lower-level struct usage to higher-level units; and (ii) Rust-oriented annotations which encode ownership, mutability, nullability, and lifetime. Synthesizing the \kg{} with LLMs, we further propose \ourtool{}, which implements a project-level C-to-Rust translation technique. In \ourtool{}, the \kg{} provides LLMs with comprehensive pointer semantics from a global perspective, thus guiding LLMs towards generating safe and idiomatic Rust code from a given C project. Our experiments show that \ourtool{} reduces unsafe usages in translated Rust by 99.9\% compared to both rule-based translation and traditional LLM-based rewriting, while achieving an average 29.3\% higher functional correctness than those fuzzing-enhanced LLM methods.

The capacity to predict human spatial preferences within built environments is instrumental for developing Cyber-Physical-Social Infrastructure Systems (CPSIS). A significant challenge in this domain is the generalizability of preference models, particularly their efficacy in predicting preferences within environmental configurations not encountered during training. While deep learning models have shown promise in learning complex spatial and contextual dependencies, it remains unclear which neural network architectures are most effective at generalizing to unseen layouts. To address this, we conduct a comparative study of Graph Neural Networks, Convolutional Neural Networks, and standard feedforward Neural Networks using synthetic data generated from a simplified and synthetic pocket park environment. Beginning with this illustrative case study, allows for controlled analysis of each model's ability to transfer learned preference patterns to unseen spatial scenarios. The models are evaluated based on their capacity to predict preferences influenced by heterogeneous physical, environmental, and social features. Generalizability score is calculated using the area under the precision-recall curve for the seen and unseen layouts. This generalizability score is appropriate for imbalanced data, providing insights into the suitability of each neural network architecture for preference-aware human behavior modeling in unseen built environments.

Early detection of Alzheimer's disease (AD) is crucial because its neurodegenerative effects are irreversible, and neuropathologic and social-behavioral risk factors accumulate years before diagnosis. Identifying higher-risk individuals earlier enables prevention, timely care, and equitable resource allocation. We predict cognitive performance from social determinants of health (SDOH) using the NIH NIA-supported PREPARE Challenge Phase 2 dataset derived from the nationally representative Mex-Cog cohort of the 2003 and 2012 Mexican Health and Aging Study (MHAS). Data: The target is a validated composite cognitive score across seven domains-orientation, memory, attention, language, constructional praxis, and executive function-derived from the 2016 and 2021 MHAS waves. Predictors span demographic, socioeconomic, health, lifestyle, psychosocial, and healthcare access factors. Methodology: Missingness was addressed with a singular value decomposition (SVD)-based imputation pipeline treating continuous and categorical variables separately. This approach leverages latent feature correlations to recover missing values while balancing reliability and scalability. After evaluating multiple methods, XGBoost was chosen for its superior predictive performance. Results and Discussion: The framework outperformed existing methods and the data challenge leaderboard, demonstrating high accuracy, robustness, and interpretability. SHAP-based post hoc analysis identified top contributing SDOH factors and age-specific feature patterns. Notably, flooring material emerged as a strong predictor, reflecting socioeconomic and environmental disparities. Other influential factors, age, SES, lifestyle, social interaction, sleep, stress, and BMI, underscore the multifactorial nature of cognitive aging and the value of interpretable, data-driven SDOH modeling.

Bias in large language models (LLMs) remains a persistent challenge, manifesting in stereotyping and unfair treatment across social groups. While prior research has primarily focused on individual models, the rise of multi-agent systems (MAS), where multiple LLMs collaborate and communicate, introduces new and largely unexplored dynamics in bias emergence and propagation. In this work, we present a comprehensive study of stereotypical bias in MAS, examining how internal specialization, underlying LLMs and inter-agent communication protocols influence bias robustness, propagation, and amplification. We simulate social contexts where agents represent different social groups and evaluate system behavior under various interaction and adversarial scenarios. Experiments on three bias benchmarks reveal that MAS are generally less robust than single-agent systems, with bias often emerging early through in-group favoritism. However, cooperative and debate-based communication can mitigate bias amplification, while more robust underlying LLMs improve overall system stability. Our findings highlight critical factors shaping fairness and resilience in multi-agent LLM systems.

Modern enterprises manage vast knowledge distributed across heterogeneous systems such as Jira, Git repositories, Confluence, and wikis. Conventional retrieval methods based on keyword search or static embeddings often fail to answer complex queries that require contextual reasoning and multi-hop inference across artifacts. We present a modular hybrid retrieval framework for adaptive enterprise information access that integrates Knowledge Base Language-Augmented Models (KBLam), DeepGraph representations, and embedding-driven semantic search. The framework builds a unified knowledge graph from parsed repositories including code, pull requests, and commit histories, enabling semantic similarity search, structural inference, and multi-hop reasoning. Query analysis dynamically determines the optimal retrieval strategy, supporting both structured and unstructured data sources through independent or fused processing. An interactive interface provides graph visualizations, subgraph exploration, and context-aware query routing to generate concise and explainable answers. Experiments on large-scale Git repositories show that the unified reasoning layer improves answer relevance by up to 80 percent compared with standalone GPT-based retrieval pipelines. By combining graph construction, hybrid reasoning, and interactive visualization, the proposed framework offers a scalable, explainable, and user-centric foundation for intelligent knowledge assistants in enterprise environments.

Reinforcement learning (RL) has been an important machine learning paradigm for solving long-horizon sequential decision-making problems under uncertainty. By integrating deep neural networks (DNNs) into the RL framework, deep reinforcement learning (DRL) has emerged, which achieved significant success in various domains. However, the integration of DNNs also makes it vulnerable to adversarial attacks. Existing adversarial attack techniques mainly focus on either directly manipulating the environment with which a victim agent interacts or deploying an adversarial agent that interacts with the victim agent to induce abnormal behaviors. While these techniques achieve promising results, their adoption in multi-party open systems remains limited due to two major reasons: impractical assumption of full control over the environment and dependent on interactions with victim agents. To enable adversarial attacks in multi-party open systems, in this paper, we redesigned an adversarial policy learning approach that can mislead well-trained victim agents without requiring direct interactions with these agents or full control over their environments. Particularly, we propose a neutral agent-based approach across various task scenarios in multi-party open systems. While the neutral agents seemingly are detached from the victim agents, indirectly influence them through the shared environment. We evaluate our proposed method on the SMAC platform based on Starcraft II and the autonomous driving simulation platform Highway-env. The experimental results demonstrate that our method can launch general and effective adversarial attacks in multi-party open systems.

We present a reproducibility study of the state-of-the-art neural architecture for sequence labeling proposed by Ma and Hovy (2016)\cite{ma2016end}. The original BiLSTM-CNN-CRF model combines character-level representations via Convolutional Neural Networks (CNNs), word-level context modeling through Bi-directional Long Short-Term Memory networks (BiLSTMs), and structured prediction using Conditional Random Fields (CRFs). This end-to-end approach eliminates the need for hand-crafted features while achieving excellent performance on named entity recognition (NER) and part-of-speech (POS) tagging tasks. Our implementation successfully reproduces the key results, achieving 91.18\% F1-score on CoNLL-2003 NER and demonstrating the model's effectiveness across sequence labeling tasks. We provide a detailed analysis of the architecture components and release an open-source PyTorch implementation to facilitate further research.

With the growing deployment of Vision-Language-Action (VLA) models in real-world embodied AI systems, their increasing vulnerability to backdoor attacks poses a serious safety threat. A backdoored VLA agent can be covertly triggered by a pre-injected backdoor to execute adversarial actions, potentially causing system failures or even physical harm. Although backdoor attacks on VLA models have been explored, prior work has focused only on untargeted attacks, leaving the more practically threatening scenario of targeted manipulation unexamined. In this paper, we study targeted backdoor attacks on VLA models and introduce TabVLA, a novel framework that enables such attacks via black-box fine-tuning. TabVLA explores two deployment-relevant inference-time threat models: input-stream editing and in-scene triggering. It formulates poisoned data generation as an optimization problem to improve attack effectivess. Experiments with OpenVLA-7B on the LIBERO benchmark reveal that the vision channel is the principal attack surface: targeted backdoors succeed with minimal poisoning, remain robust across variations in trigger design, and are degraded only by positional mismatches between fine-tuning and inference triggers. We also investigate a potential detection-based defense against TabVLA, which reconstructs latent visual triggers from the input stream to flag activation-conditioned backdoor samples. Our work highlights the vulnerability of VLA models to targeted backdoor manipulation and underscores the need for more advanced defenses.

Reasoning is not just about solving problems -- it is also about evaluating which problems are worth solving at all. Evaluations of artificial intelligence (AI) systems primarily focused on problem solving, historically by studying how models play games such as chess and Go. In this paper, we advocate for a new paradigm that assesses AI systems' evaluation of games. First, we introduce a formalism for evaluating such evaluations. We then leverage a large-scale dataset of over $100$ novel board games and over 450 human judgments to compare evaluations produced by modern language and reasoning models against those of people and symbolic computational agents. We consider two kinds of evaluative queries: assessing the payoff (or fairness) and the funness of games. These queries span two dimensions relevant to the design of evaluations of AI evaluations: how complex a query is to compute and how difficult a query is to quantify. Our results show that reasoning models are generally more aligned to people in their evaluations of games than non-reasoning language models. However, we observe a non-monotonic relationship: as models get closer to game-theoretic optimal, their fit to human data weakens. We also observe more "jaggedness" across models for assessing funness, in line with the greater difficulty of quantifying this query. Across queries and games, reasoning models show highly variable and unpredictable resource usage when assessing queries, pointing to the importance of imbuing more resource-rational meta-reasoning in language and reasoning models.