We provide the first proof of learning rate transfer with width in a linear multi-layer perceptron (MLP) parametrized with $μ$P, a neural network parameterization designed to ``maximize'' feature learning in the infinite-width limit. We show that under $μP$, the optimal learning rate converges to a \emph{non-zero constant} as width goes to infinity, providing a theoretical explanation to learning rate transfer. In contrast, we show that this property fails to hold under alternative parametrizations such as Standard Parametrization (SP) and Neural Tangent Parametrization (NTP). We provide intuitive proofs and support the theoretical findings with extensive empirical results.

Large Language Model (LLM) deployment requires guiding the LLM to recognize and not answer unsafe prompts while complying with safe prompts. Previous methods for achieving this require adjusting model weights along with other expensive procedures. While recent advances in Sparse Autoencoders (SAEs) have enabled interpretable feature extraction from LLMs, existing approaches lack systematic feature selection methods and principled evaluation of safety-utility tradeoffs. We explored using different steering features and steering strengths using Sparse Auto Encoders (SAEs) to provide a solution. Using an accurate and innovative contrasting prompt method with the AI-Generated Prompts Dataset from teknium/OpenHermes-2p5-Mistral-7B and Air Bench eu-dataset to efficiently choose the best features in the model to steer, we tested this method on Llama-3 8B. We conclude that using this method, our approach achieves an 18.9% improvement in safety performance while simultaneously increasing utility by 11.1%, demonstrating that targeted SAE steering can overcome traditional safety-utility tradeoffs when optimal features are identified through principled selection methods.

Manual annotation of volumetric medical images, such as magnetic resonance imaging (MRI) and computed tomography (CT), is a labor-intensive and time-consuming process. Recent advancements in foundation models for video object segmentation, such as Segment Anything Model 2 (SAM 2), offer a potential opportunity to significantly speed up the annotation process by manually annotating one or a few slices and then propagating target masks across the entire volume. However, the performance of SAM 2 in this context varies. Our experiments show that relying on a single memory bank and attention module is prone to error propagation, particularly at boundary regions where the target is present in the previous slice but absent in the current one. To address this problem, we propose Short-Long Memory SAM 2 (SLM-SAM 2), a novel architecture that integrates distinct short-term and long-term memory banks with separate attention modules to improve segmentation accuracy. We evaluate SLM-SAM 2 on four public datasets covering organs, bones, and muscles across MRI, CT, and ultrasound videos. We show that the proposed method markedly outperforms the default SAM 2, achieving an average Dice Similarity Coefficient improvement of 0.14 and 0.10 in the scenarios when 5 volumes and 1 volume are available for the initial adaptation, respectively. SLM-SAM 2 also exhibits stronger resistance to over-propagation, reducing the time required to correct propagated masks by 60.575% per volume compared to SAM 2, making a notable step toward more accurate automated annotation of medical images for segmentation model development.

Large language models (LLMs) increasingly mediate decisions in domains where unfair treatment of demographic groups is unacceptable. Existing work probes when biased outputs appear, but gives little insight into the mechanisms that generate them, leaving existing mitigations largely fragile. In this paper, we conduct a systematic investigation LLM unfairness and propose DiffHeads, a lightweight debiasing framework for LLMs. We first compare Direct-Answer (DA) prompting to Chain-of-Thought (CoT) prompting across eight representative open- and closed-source LLMs. DA will trigger the nature bias part of LLM and improve measured unfairness by 534.5%-391.9% in both one-turn and two-turn dialogues. Next, we define a token-to-head contribution score that traces each token's influence back to individual attention heads. This reveals a small cluster of bias heads that activate under DA but stay largely dormant with CoT, providing the first causal link between prompting strategy and bias emergence. Finally, building on this insight, we propose DiffHeads that identifies bias heads through differential activation analysis between DA and CoT, and selectively masks only those heads. DiffHeads reduces unfairness by 49.4%, and 40.3% under DA and CoT, respectively, without harming model utility.

Tractography fiber clustering using diffusion MRI (dMRI) is a crucial method for white matter (WM) parcellation to enable analysis of brains structural connectivity in health and disease. Current fiber clustering strategies primarily use the fiber geometric characteristics (i.e., the spatial trajectories) to group similar fibers into clusters, while neglecting the functional and microstructural information of the fiber tracts. There is increasing evidence that neural activity in the WM can be measured using functional MRI (fMRI), providing potentially valuable multimodal information for fiber clustering to enhance its functional coherence. Furthermore, microstructural features such as fractional anisotropy (FA) can be computed from dMRI as additional information to ensure the anatomical coherence of the clusters. In this paper, we develop a novel deep learning fiber clustering framework, namely Deep Multi-view Fiber Clustering (DMVFC), which uses joint multi-modal dMRI and fMRI data to enable functionally consistent WM parcellation. DMVFC can effectively integrate the geometric and microstructural characteristics of the WM fibers with the fMRI BOLD signals along the fiber tracts. DMVFC includes two major components: (1) a multi-view pretraining module to compute embedding features from each source of information separately, including fiber geometry, microstructure measures, and functional signals, and (2) a collaborative fine-tuning module to simultaneously refine the differences of embeddings. In the experiments, we compare DMVFC with two state-of-the-art fiber clustering methods and demonstrate superior performance in achieving functionally meaningful and consistent WM parcellation results.

Large language models (LLMs) have demonstrated that large-scale pretraining enables systems to adapt rapidly to new problems with little supervision in the language domain. This success, however, has not translated as effectively to the visual domain, where models, including LLMs, continue to struggle with compositional understanding, sample efficiency, and general-purpose problem-solving. We investigate Video Diffusion Models (VDMs) as a promising direction for bridging this gap. Pretraining on spatiotemporal data endows these models with strong inductive biases for structure and dynamics, which we hypothesize can support broad task adaptability. To test this, we design a controlled evaluation in which both a pretrained LLM and a pretrained VDM are equipped with lightweight adapters and presented with tasks in their natural modalities. Across benchmarks including ARC-AGI, ConceptARC, visual games, route planning, and cellular automata, VDMs demonstrate higher data efficiency than their language counterparts. Taken together, our results indicate that video pretraining offers inductive biases that support progress toward visual foundation models.

Ensuring that Large Language Models (LLMs) align with the diverse and evolving human values across different regions and cultures remains a critical challenge in AI ethics. Current alignment approaches often yield superficial conformity rather than genuine ethical understanding, failing to address the complex, context-dependent nature of human values. In this paper, we propose a novel ethical reasoning paradigm for LLMs inspired by well-established ethical decision-making models, aiming at enhancing diverse human value alignment through deliberative ethical reasoning. Our framework consists of a structured five-step process, including contextual fact gathering, hierarchical social norm identification, option generation, multiple-lens ethical impact analysis, and reflection. This theory-grounded approach guides LLMs through an interpretable reasoning process that enhances their ability to understand regional specificities and perform nuanced ethical analysis, which can be implemented with either prompt engineering or supervised fine-tuning methods. We perform evaluations on the SafeWorld benchmark that specially designed for regional value alignment. Experimental results demonstrate our framework significantly improves LLM alignment with diverse human values compared to baseline methods, enabling more accurate social norm identification and more culturally appropriate reasoning. Our work provides a concrete pathway toward developing LLMs that align more effectively with the multifaceted values of global societies through interdisciplinary research.

The advances and availability of technologies involving Generative Artificial Intelligence (AI) are evolving clearly and explicitly, driving immediate changes in various work activities. Software Engineering (SE) is no exception and stands to benefit from these new technologies, enhancing productivity and quality in its software development processes. However, although the use of Generative AI in SE practices is still in its early stages, considering the lack of conclusive results from ongoing research and the limited technological maturity, we have chosen to incorporate these technologies in the development of a web-based software system to be used in clinical trials by a thoracic diseases research group at our university. For this reason, we decided to share this experience report documenting our development team's learning journey in using Generative AI during the software development process. Project management, requirements specification, design, development, and quality assurance activities form the scope of observation. Although we do not yet have definitive technological evidence to evolve our development process significantly, the results obtained and the suggestions shared here represent valuable insights for software organizations seeking to innovate their development practices to achieve software quality with generative AI.

Domain adaptation is required for automated driving models to generalize well across diverse road conditions. This paper explores a training method for domain adaptation to adapt PilotNet, an end-to-end deep learning-based model, for left-hand driving conditions using real-world Australian highway data. Four training methods were evaluated: (1) a baseline model trained on U.S. right-hand driving data, (2) a model trained on flipped U.S. data, (3) a model pretrained on U.S. data and then fine-tuned on Australian highways, and (4) a model pretrained on flipped U.S. data and then finetuned on Australian highways. This setup examines whether incorporating flipped data enhances the model adaptation by providing an initial left-hand driving alignment. The paper compares model performance regarding steering prediction accuracy and attention, using saliency-based analysis to measure attention shifts across significant road regions. Results show that pretraining on flipped data alone worsens prediction stability due to misaligned feature representations, but significantly improves adaptation when followed by fine-tuning, leading to lower prediction error and stronger focus on left-side cues. To validate this approach across different architectures, the same experiments were done on ResNet, which confirmed similar adaptation trends. These findings emphasize the importance of preprocessing techniques, such as flipped-data pretraining, followed by fine-tuning to improve model adaptation with minimal retraining requirements.

Retrieval-Augmented Generation (RAG) quality depends on many interacting choices across retrieval, ranking, augmentation, prompting, and generation, so optimizing modules in isolation is brittle. We introduce RAGSmith, a modular framework that treats RAG design as an end-to-end architecture search over nine technique families and 46{,}080 feasible pipeline configurations. A genetic search optimizes a scalar objective that jointly aggregates retrieval metrics (recall@k, mAP, nDCG, MRR) and generation metrics (LLM-Judge and semantic similarity). We evaluate on six Wikipedia-derived domains (Mathematics, Law, Finance, Medicine, Defense Industry, Computer Science), each with 100 questions spanning factual, interpretation, and long-answer types. RAGSmith finds configurations that consistently outperform naive RAG baseline by +3.8\% on average (range +1.2\% to +6.9\% across domains), with gains up to +12.5\% in retrieval and +7.5\% in generation. The search typically explores $\approx 0.2\%$ of the space ($\sim 100$ candidates) and discovers a robust backbone -- vector retrieval plus post-generation reflection/revision -- augmented by domain-dependent choices in expansion, reranking, augmentation, and prompt reordering; passage compression is never selected. Improvement magnitude correlates with question type, with larger gains on factual/long-answer mixes than interpretation-heavy sets. These results provide practical, domain-aware guidance for assembling effective RAG systems and demonstrate the utility of evolutionary search for full-pipeline optimization.