While recent advances in preference learning have enhanced alignment in human feedback, mathematical reasoning remains a persistent challenge. We investigate how data diversification strategies in preference optimization can improve the mathematical reasoning abilities of large language models (LLMs). We evaluate three common data generation methods: temperature sampling, Chain-of-Thought prompting, and Monte Carlo Tree Search (MCTS), and introduce Diversified-ThinkSolve (DTS), a novel structured approach that systematically decomposes problems into diverse reasoning paths. Our results show that with strategically diversified preference data, models can substantially improve mathematical reasoning performance, with the best approach yielding gains of 7.1% on GSM8K and 4.2% on MATH over the base model. Despite its strong performance, DTS incurs only a marginal computational overhead (1.03x) compared to the baseline, while MCTS is nearly five times more costly with lower returns. These findings demonstrate that structured exploration of diverse problem-solving methods creates more effective preference data for mathematical alignment than traditional approaches.

When Jocelyn Leitzinger had her university students write about times in their lives when they had witnessed discrimination, she noticed that a woman named Sally was the victim in many of the stories. "It was very clear that ChatGPT had decided this is a common woman's name," said Leitzinger, who teaches an undergraduate class on business and society at the University of Illinois in Chicago. "They weren't even coming up with their own anecdotal stories about their own lives," she said. Leitzinger estimated that around half of her 180 students used ChatGPT inappropriately at some point last semester -- including when writing about the ethics of artificial intelligence (AI), which she called both "ironic" and "mind-boggling."

In this letter, we present tightly coupled LiDAR-IMU-leg odometry, which is robust to challenging conditions such as featureless environments and deformable terrains. We developed an online learning-based leg kinematics model named the neural leg kinematics model, which incorporates tactile information (foot reaction force) to implicitly express the nonlinear dynamics between robot feet and the ground. Online training of this model enhances its adaptability to weight load changes of a robot (e.g., assuming delivery or transportation tasks) and terrain conditions. According to the \textit{neural adaptive leg odometry factor} and online uncertainty estimation of the leg kinematics model-based motion predictions, we jointly solve online training of this kinematics model and odometry estimation on a unified factor graph to retain the consistency of both. The proposed method was verified through real experiments using a quadruped robot in two challenging situations: 1) a sandy beach, representing an extremely featureless area with a deformable terrain, and 2) a campus, including multiple featureless areas and terrain types of asphalt, gravel (deformable terrain), and grass. Experimental results showed that our odometry estimation incorporating the \textit{neural leg kinematics model} outperforms state-of-the-art works. Our project page is available for further details: https://takuokawara.github.io/RAL2025_project_page/

Recent work in continual learning has highlighted the beneficial effect of resampling weights in the last layer of a neural network ("zapping"). Although empirical results demonstrate the effectiveness of this approach, the underlying mechanisms that drive these improvements remain unclear. In this work, we investigate in detail the pattern of learning and forgetting that take place inside a convolutional neural network when trained in challenging settings such as continual learning and few-shot transfer learning, with handwritten characters and natural images. Our experiments show that models that have undergone zapping during training more quickly recover from the shock of transferring to a new domain. Furthermore, to better observe the effect of continual learning in a multi-task setting we measure how each individual task is affected. This shows that, not only zapping, but the choice of optimizer can also deeply affect the dynamics of learning and forgetting, causing complex patterns of synergy/interference between tasks to emerge when the model learns sequentially at transfer time. Despite the popularity of deep learning, neural network training is still largely considered a "dark art" (Lee et al., 2020). This alchemical connotation is in no small part due to the difficulty of building reliable intuitions about optimization in high-dimensional spaces. Analysis of neural network loss landscapes Li et al. (2018b)--the map of a network's weights to their corresponding loss values--reveals which training mechanisms are effective and helps develop new methods that account for the landscape structure. Our work builds upon the work of Javed & White (2019); Beaulieu et al. (2020); Frati et al. (2024) and sheds new light on the effect of resampling weights during pre-training, and the dynamics of learning and forgetting while navigating the complex loss landscapes of transfer (Zhuang et al., 2020) and continual (Wang et al., 2024) learning problems.

We introduce a simple yet effective technique for estimating lighting from a single low-dynamic-range (LDR) image by reframing the task as a chrome ball inpainting problem. This approach leverages a pre-trained diffusion model, Stable Diffusion XL, to overcome the generalization failures of existing methods that rely on limited HDR panorama datasets. While conceptually simple, the task remains challenging because diffusion models often insert incorrect or inconsistent content and cannot readily generate chrome balls in HDR format. Our analysis reveals that the inpainting process is highly sensitive to the initial noise in the diffusion process, occasionally resulting in unrealistic outputs. To address this, we first introduce DiffusionLight, which uses iterative inpainting to compute a median chrome ball from multiple outputs to serve as a stable, low-frequency lighting prior that guides the generation of a high-quality final result. To generate high-dynamic-range (HDR) light probes, an Exposure LoRA is fine-tuned to create LDR images at multiple exposure values, which are then merged. While effective, DiffusionLight is time-intensive, requiring approximately 30 minutes per estimation. To reduce this overhead, we introduce DiffusionLight-Turbo, which reduces the runtime to about 30 seconds with minimal quality loss. This 60x speedup is achieved by training a Turbo LoRA to directly predict the averaged chrome balls from the iterative process. Inference is further streamlined into a single denoising pass using a LoRA swapping technique. Experimental results that show our method produces convincing light estimates across diverse settings and demonstrates superior generalization to in-the-wild scenarios. Our code is available at https://diffusionlight.github.io/turbo

Despite the recent strides in large language models, studies have underscored the existence of social biases within these systems. In this paper, we delve into the validation and comparison of the ethical biases of LLMs concerning globally discussed and potentially sensitive topics, hypothesizing that these biases may arise from language-specific distinctions. Introducing the Multilingual Sensitive Questions & Answers Dataset (MSQAD), we collected news articles from Human Rights Watch covering 17 topics, and generated socially sensitive questions along with corresponding responses in multiple languages. We scrutinized the biases of these responses across languages and topics, employing two statistical hypothesis tests. The results showed that the null hypotheses were rejected in most cases, indicating biases arising from cross-language differences. It demonstrates that ethical biases in responses are widespread across various languages, and notably, these biases were prevalent even among different LLMs. By making the proposed MSQAD openly available, we aim to facilitate future research endeavors focused on examining cross-language biases in LLMs and their variant models.

This study presents a novel classroom surveillance system that integrates multiple modalities, including drowsiness, tracking of mobile phone usage, and face recognition,to assess student attentiveness with enhanced precision.The system leverages the YOLOv8 model to detect both mobile phone and sleep usage,(Ghatge et al., 2024) while facial recognition is achieved through LResNet Occ FC body tracking using YOLO and MTCNN.(Durai et al., 2024) These models work in synergy to provide comprehensive, real-time monitoring, offering insights into student engagement and behavior.(S et al., 2023) The framework is trained on specialized datasets, such as the RMFD dataset for face recognition and a Roboflow dataset for mobile phone detection. The extensive evaluation of the system shows promising results. Sleep detection achieves 97. 42% mAP@50, face recognition achieves 86. 45% validation accuracy and mobile phone detection reach 85. 89% mAP@50. The system is implemented within a core PHP web application and utilizes ESP32-CAM hardware for seamless data capture.(Neto et al., 2024) This integrated approach not only enhances classroom monitoring, but also ensures automatic attendance recording via face recognition as students remain seated in the classroom, offering scalability for diverse educational environments.(Banada,2025)

Time series forecasting models have diverse real world applications (e.g., from electricity metrics to software workload). Latest foundational models trained for time series forecasting show strengths (e.g., for long sequences and in zero-shot settings). However, foundational model was not yet used for forecasting rare, spiky events, i.e., a challenging target because those are a corner case of extreme events. In this paper, we optimize a state-of-the-art foundational model to forecast sporadic or spiky production outages of high-performance machine learning services powering billions of client devices. We evaluate the forecasting errors of the foundational model compared with classical stochastic forecasting models (e.g., moving average and autoregressive). The analysis helps us understand how each of the evaluated models performs for the sporadic or spiky events. For example, it identifies the key patterns in the target data that are well tracked by the foundational model vs. each of the stochastic models. We use the models with optimal parameters to estimate a year-long outage statistics of a particular root cause with less than 6% value errors.

Large language models (LLMs) excel at reasoning tasks requiring long thought sequences for planning, reflection, and refinement. However, their substantial model size and high computational demands are impractical for widespread deployment. Yet, small language models (SLMs) often struggle to learn long-form CoT reasoning due to their limited capacity, a phenomenon we refer to as the "SLMs Learnability Gap". To address this, we introduce \textbf{Mi}d-\textbf{Co}T \textbf{T}eacher \textbf{A}ssistant Distillation (MiCoTAl), a framework for improving long CoT distillation for SLMs. MiCoTA employs intermediate-sized models as teacher assistants and utilizes intermediate-length CoT sequences to bridge both the capacity and reasoning length gaps. Our experiments on downstream tasks demonstrate that although SLMs distilled from large teachers can perform poorly, by applying MiCoTA, they achieve significant improvements in reasoning performance. Specifically, Qwen2.5-7B-Instruct and Qwen2.5-3B-Instruct achieve an improvement of 3.47 and 3.93 respectively on average score on AIME2024, AMC, Olympiad, MATH-500 and GSM8K benchmarks. To better understand the mechanism behind MiCoTA, we perform a quantitative experiment demonstrating that our method produces data more closely aligned with base SLM distributions. Our insights pave the way for future research into long-CoT data distillation for SLMs.

There is justifiable interest in leveraging conversational AI (CAI) for health across the majority world, but to be effective, CAI must respond appropriately within cultur ally and linguistically diverse context s . Therefore, we need ways to address the fact that current LLMs exclude many lived experience s globally . Various advances are underway which focus on top - down approaches and increas ing training data . In this paper, we aim to complement these with a bottom - up locally - grounded approach based on qualitative data collected during participatory workshops in Latin America. Our goal is to construct a rich and human - centred understanding o f: a) potential areas of cultural misalignment in digital health; b) regional perspectives on chatbots for health and c) strategies for creating culturally - appropriate CAI; with a focus on the understudied Latin American context . Our findings show that academic boundaries on notions of cultur e lose meaning at the ground level and technologies will need to engage with a broad er framework; one that encapsulates the way economics, politics, geogr aphy and local logistics are entangled in cultural experience. To this end, we introduce a framework for ' Pluriversal Conversational AI for H ealth ' which allows for the possibility that more relationality and tolerance, rather than just more data, may be called for .