While large general-purpose Transformer-based encoders excel at general language understanding, their performance diminishes in specialized domains like manufacturing due to a lack of exposure to domain-specific terminology and semantics. In this paper, we address this gap by introducing ManufactuBERT, a RoBERTa model continually pretrained on a large-scale corpus curated for the manufacturing domain. We present a comprehensive data processing pipeline to create this corpus from web data, involving an initial domain-specific filtering step followed by a multi-stage deduplication process that removes redundancies. Our experiments show that ManufactuBERT establishes a new state-of-the-art on a range of manufacturing-related NLP tasks, outperforming strong specialized baselines. More importantly, we demonstrate that training on our carefully deduplicated corpus significantly accelerates convergence, leading to a 33\% reduction in training time and computational cost compared to training on the non-deduplicated dataset. The proposed pipeline offers a reproducible example for developing high-performing encoders in other specialized domains. We will release our model and curated corpus at https://huggingface.co/cea-list-ia.

The loss function used to train a neural network is strongly connected to its output layer from a statistical point of view. This technical report analyzes common activation functions for a neural network output layer, like linear, sigmoid, ReLU, and softmax, detailing their mathematical properties and their appropriate use cases. A strong statistical justification exists for the selection of the suitable loss function for training a deep learning model. This report connects common loss functions such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and various Cross-Entropy losses to the statistical principle of Maximum Likelihood Estimation (MLE). Choosing a specific loss function is equivalent to assuming a specific probability distribution for the model output, highlighting the link between these functions and the Generalized Linear Models (GLMs) that underlie network output layers. Additional scenarios of practical interest are also considered, such as alternative output encodings, constrained outputs, and distributions with heavy tails.

Text offers intuitive access to information. This can, in particular, complement the density of numerical time series, thereby allowing improved interactions with time series models to enhance accessibility and decision-making. While the creation of question-answering datasets and models has recently seen remarkable growth, most research focuses on question answering (QA) on vision and text, with time series receiving minute attention. To bridge this gap, we propose a challenging novel time series QA (TSQA) dataset, QuAnTS, for Question Answering on Time Series data. Specifically, we pose a wide variety of questions and answers about human motion in the form of tracked skeleton trajectories. We verify that the large-scale QuAnTS dataset is well-formed and comprehensive through extensive experiments. Thoroughly evaluating existing and newly proposed baselines then lays the groundwork for a deeper exploration of TSQA using QuAnTS. Additionally, we provide human performances as a key reference for gauging the practical usability of such models. We hope to encourage future research on interacting with time series models through text, enabling better decision-making and more transparent systems.

Creating programs to represent board games can be a time-consuming task. Large Language Models (LLMs) arise as appealing tools to expedite this process, given their capacity to efficiently generate code from simple contextual information. In this work, we propose a method to test how capable three LLMs (Claude, DeepSeek and ChatGPT) are at creating code for board games, as well as new variants of existing games.

Alterations in retinal layer thickness, measurable using Optical Coherence Tomography (OCT), have been associated with neurodegenerative diseases such as Alzheimer's disease (AD). While previous studies have mainly focused on segmented layer thickness measurements, this study explored the direct classification of OCT B-scan images for the early detection of AD. To our knowledge, this is the first application of deep learning to raw OCT B-scans for AD prediction in the literature. Unlike conventional medical image classification tasks, early detection is more challenging than diagnosis because imaging precedes clinical diagnosis by several years. We fine-tuned and evaluated multiple pretrained models, including ImageNet-based networks and the OCT-specific RETFound transformer, using subject-level cross-validation datasets matched for age, sex, and imaging instances from the UK Biobank cohort. To reduce overfitting in this small, high-dimensional dataset, both standard and OCT-specific augmentation techniques were applied, along with a year-weighted loss function that prioritized cases diagnosed within four years of imaging. ResNet-34 produced the most stable results, achieving an AUC of 0.62 in the 4-year cohort. Although below the threshold for clinical application, our explainability analyses confirmed localized structural differences in the central macular subfield between the AD and control groups. These findings provide a baseline for OCT-based AD prediction, highlight the challenges of detecting subtle retinal biomarkers years before AD diagnosis, and point to the need for larger datasets and multimodal approaches.

This work investigates distillation methods for large language models (LLMs) with the goal of developing compact models that preserve high performance. Several existing approaches are reviewed, with a discussion of their respective strengths and limitations. An improved method based on the ShortGPT approach has been developed, building upon the idea of incorporating iterative evaluation of layer importance. At each step, importance is assessed by measuring performance degradation when individual layers are removed, using a set of representative datasets. This process is combined with further training using a joint loss function based on KL divergence and mean squared error. Experiments on the Qwen2.5-3B model show that the number of layers can be reduced from 36 to 28 (resulting in a 2.47 billion parameter model) with only a 9.7% quality loss, and to 24 layers with an 18% loss. The findings suggest that the middle transformer layers contribute less to inference, underscoring the potential of the proposed method for creating efficient models. The results demonstrate the effectiveness of iterative distillation and fine-tuning, making the approach suitable for deployment in resource-limited settings.

In this paper, we present a novel series of Russian information retrieval datasets constructed from the "Did you know..." section of Russian Wikipedia. Our datasets support a range of retrieval tasks, including fact-checking, retrieval-augmented generation, and full-document retrieval, by leveraging interesting facts and their referenced Wikipedia articles annotated at the sentence level with graded relevance. We describe the methodology for dataset creation that enables the expansion of existing Russian Information Retrieval (IR) resources. Through extensive experiments, we extend the RusBEIR research by comparing lexical retrieval models, such as BM25, with state-of-the-art neural architectures fine-tuned for Russian, as well as multilingual models. Results of our experiments show that lexical methods tend to outperform neural models on full-document retrieval, while neural approaches better capture lexical semantics in shorter texts, such as in fact-checking or fine-grained retrieval. Using our newly created datasets, we also analyze the impact of document length on retrieval performance and demonstrate that combining retrieval with neural reranking consistently improves results. Our contribution expands the resources available for Russian information retrieval research and highlights the importance of accurate evaluation of retrieval models to achieve optimal performance. All datasets are publicly available at HuggingFace. To facilitate reproducibility and future research, we also release the full implementation on GitHub.

Understanding intrinsic differences between adversarial examples and clean samples is key to enhancing DNN robustness and detection against adversarial attacks. This study first empirically finds that image-based adversarial examples are notably sensitive to occlusion. Controlled experiments on CIFAR-10 used nine canonical attacks (e.g., FGSM, PGD) to generate adversarial examples, paired with original samples for evaluation. We introduce Sliding Mask Confidence Entropy (SMCE) to quantify model confidence fluctuation under occlusion. Using 1800+ test images, SMCE calculations supported by Mask Entropy Field Maps and statistical distributions show adversarial examples have significantly higher confidence volatility under occlusion than originals. Based on this, we propose Sliding Window Mask-based Adversarial Example Detection (SWM-AED), which avoids catastrophic overfitting of conventional adversarial training. Evaluations across classifiers and attacks on CIFAR-10 demonstrate robust performance, with accuracy over 62% in most cases and up to 96.5%.

In this paper, we explore an important yet previously neglected question: Do context aggregation patterns across Language Models (LMs) share commonalities? While some works have investigated context aggregation or attention weights in LMs, they typically focus on individual models or attention heads, lacking a systematic analysis across multiple LMs to explore their commonalities. In contrast, we focus on the commonalities among LMs, which can deepen our understanding of LMs and even facilitate cross-model knowledge transfer. In this work, we introduce the Order-Level Attention (OLA) derived from the order-wise decomposition of Attention Rollout and reveal that the OLA at the same order across LMs exhibits significant similarities. Furthermore, we discover an implicit mapping between OLA and syntactic knowledge. Based on these two findings, we propose the Transferable OLA Adapter (TOA), a training-free cross-LM adapter transfer method. Specifically, we treat the OLA as a unified syntactic feature representation and train an adapter that takes OLA as input. Due to the similarities in OLA across LMs, the adapter generalizes to unseen LMs without requiring any parameter updates. Extensive experiments demonstrate that TOA's cross-LM generalization effectively enhances the performance of unseen LMs. Code is available at https://github.com/jinglin-liang/OLAS.

Machine learning based on neural networks has advanced rapidly, but the high energy consumption required for training and inference remains a major challenge. Hyperdimensional Computing (HDC) offers a lightweight, brain-inspired alternative that enables high parallelism but often suffers from lower accuracy on complex visual tasks. To overcome this, hybrid accelerators combining HDC and Convolutional Neural Networks (CNNs) have been proposed, though their adoption is limited by poor generalizability and programmability. The rise of open-source RISC-V architectures has created new opportunities for domain-specific GPU design. Unlike traditional proprietary GPUs, emerging RISC-V-based GPUs provide flexible, programmable platforms suitable for custom computation models such as HDC. In this study, we design and implement custom GPU instructions optimized for HDC operations, enabling efficient processing for hybrid HDC-CNN workloads. Experimental results using four types of custom HDC instructions show a performance improvement of up to 56.2 times in microbenchmark tests, demonstrating the potential of RISC-V GPUs for energy-efficient, high-performance computing.