Data Science and Engineering Lab Universit e libre de Bruxelles Brussels, Belgium gaspard.merten@ulb.be Data Science and Engineering Lab Universit e libre de Bruxelles Brussels, Belgium gilles.dejaegere@ulb.be Data Science and Engineering Lab Universit e libre de Bruxelles Brussels, Belgium mahmoud.sakr@ulb.be Abstract --Recent research in Large Language Models (LLMs), has had a profound impact across various fields, including mobility data science. This paper explores the and experiment with different approaches to using LLMs for analyzing AIS data. We propose a set of carefully designed queries to assess the reasoning capabilities of LLMs in this kind of tasks. Further, we experiment with four different methods: (1) using LLMs as a natural language interface to a spatial database, (2) reasoning on raw data, (3) reasoning on compressed trajectories, and (4) reasoning on semantic trajectories. We investigate the strengths and weaknesses for the four methods, and discuss the findings. The goal is to provide valuable insights for both researchers and practitioners on selecting the most appropriate LLM-based method depending on their specific data analysis objectives. The significant development in artificial machine learning has also opened the way to new approaches to solve real-world geospatial problems. In particular, Large Language Models (LLMs) have emerged as powerful tools for understanding and generating human-like text. These models have demonstrated remarkable abilities in natural language processing tasks, from answering complex queries to summarizing and interpreting information in various domains. This exponential increase of LLMs usage can also be witnessed in the domain of Geographic Information Systems (GIS) in recent years.

--High-quality, multi-channel neural recording is indispensable for neuroscience research and clinical applications. Large-scale brain recordings often produce vast amounts of data that must be wirelessly transmitted for subsequent offline analysis and decoding, especially in brain-computer interfaces (BCIs) utilizing high-density intracortical recordings with hundreds or thousands of electrodes. However, transmitting raw neural data presents significant challenges due to limited communication bandwidth and resultant excessive heating. T o address this challenge, we propose a neural signal compression scheme utilizing Convolutional Autoencoders (CAEs), which achieves a compression ratio of up to 150 for compressing local field potentials (LFPs). The CAE encoder section is implemented on RAMAN, an energy-efficient tinyML accelerator designed for edge computing. Additionally, we employ hardware-software co-optimization by pruning the CAE encoder model parameters using a hardware-aware balanced stochastic pruning strategy, resolving workload imbalance issues and eliminating indexing overhead to reduce parameter storage requirements by up to 32.4%. Post layout simulation shows that the RAMAN encoder can be implemented in a TSMC 65-nm CMOS process, occupying a core area of 0.0187 mm Operating at a clock frequency of 2 MHz and a supply voltage of 1.2 V, the estimated power consumption is 15.1 µ W per channel for the proposed DS-CAE1 model. The compressed neural data from RAMAN is reconstructed offline with signal-to-noise and distortion ratios (SNDR) of 22.6 dB and 27.4 dB, along with R2 scores of 0.81 and 0.94, respectively, evaluated on two monkey neural recordings. A. Krishna is with the Department of Electronic Systems Engineering, Indian Institute of Science, Bangalore - 560012, India, and also with the International Centre for Neuromorphic Systems, The MARCS Institute, Western Sydney University, Australia. S. Debnath, M. Srivatsav, M. Mehendale, and C. S. Thakur (Email: csthakur@iisc.ac.in) are with the Department of Electronic Systems Engineering, Indian Institute of Science, Bangalore - 560012, India. A. van Schaik is with the International Centre for Neuromorphic Systems, The MARCS Institute, Western Sydney University, Australia. This work was supported by the Pratiksha Trust grant BCD - FG/SMCH-22-2106 and INAE grant INAE/121/AKF/48 (SAP code - SP/INAE-23-0001). BCIs have emerged as a revolutionary tool for advancing our understanding of the brain and are increasingly being utilized across various clinical applications [5]-[7], providing inventive solutions for communication [8], control [1], [9], and rehabilitation [10]-[13]. Ongoing improvements in signal processing, machine learning algorithms, and neurotechnology pave the way for BCIs to revolutionize healthcare, human-computer interaction, and beyond.

Machine unlearning is concerned with the task of removing knowledge learned from particular data points from a trained model. In the context of large language models (LLMs), unlearning has recently received increased attention, particularly for removing knowledge about named entities from models for privacy purposes. While various approaches have been proposed to address the unlearning problem, most existing approaches treat all data points to be unlearned equally, i.e., unlearning that Montreal is a city in Canada is treated exactly the same as unlearning the phone number of the first author of this paper. In this work, we show that this all data is equal assumption does not hold for LLM unlearning. We study how the success of unlearning depends on the frequency of the knowledge we want to unlearn in the pre-training data of a model and find that frequency strongly affects unlearning, i.e., more frequent knowledge is harder to unlearn. Additionally, we uncover a misalignment between probability and generation-based evaluations of unlearning and show that this problem worsens as models become larger. Overall, our experiments highlight the need for better evaluation practices and novel methods for LLM unlearning that take the training data of models into account.

Recent imitation learning policies, often framed as time series prediction tasks, directly map robotic observations into the action space, such as high-dimensional visual data and proprioception. When deploying at the edge, we found the underutilization of frequency domain analysis in robotic manipulation trajectory prediction leads to neglecting the inherent rhythm information embedded within action sequences, resulting in errors at critical moments. To address this, we reframe imitation learning policies through the lens of time-scale domain and introduce the Wavelet Policy. This novel approach employs wavelet transforms (WT) and new Features Extractor (FE) for feature preprocessing and extracts multi-scale features using the Single Encoder to Multiple Decoder (SE2MD) architecture. Furthermore, to enhance feature mapping in the scale domain and appropriately increase model capacity, we introduce a Learnable Scale Domain Filter (LSDF) after each decoder, improving adaptability under different visual conditions. Our results show that the Wavelet Policy maintaining a comparable parameter count outperforms SOTA end-to-end methods on four challenging simulation robotic arm tasks and real tasks, especially at critical moments and remote settings simultaneously. We release the source code and model checkpoint of simulation task at https://github.com/lurenjia384/Wavelet_Policy.

Preference learning is critical for aligning large language models (LLMs) with human values, yet its success hinges on high-quality datasets comprising three core components: Preference \textbf{A}nnotations, \textbf{I}nstructions, and \textbf{R}esponse Pairs. Current approaches conflate these components, obscuring their individual impacts and hindering systematic optimization. In this work, we propose \textbf{AIR}, a component-wise analysis framework that systematically isolates and optimizes each component while evaluating their synergistic effects. Through rigorous experimentation, AIR reveals actionable principles: annotation simplicity (point-wise generative scoring), instruction inference stability (variance-based filtering across LLMs), and response pair quality (moderate margins + high absolute scores). When combined, these principles yield +5.3 average gains over baseline method, even with only 14k high-quality pairs. Our work shifts preference dataset design from ad hoc scaling to component-aware optimization, offering a blueprint for efficient, reproducible alignment.

Despite being the main tool to visualize molecules at the atomic scale, AFM with CO-functionalized metal tips is unable to chemically identify the observed molecules. Here we present a strategy to address this challenging task using deep learning techniques. Instead of identifying a finite number of molecules following a traditional classification approach, we define the molecular identification as an image captioning problem. We design an architecture, composed of two multimodal recurrent neural networks, capable of identifying the structure and composition of an unknown molecule using a 3D-AFM image stack as input. The neural network is trained to provide the name of each molecule according to the IUPAC nomenclature rules. To train and test this algorithm we use the novel QUAM-AFM dataset, which contains almost 700,000 molecules and 165 million AFM images. The accuracy of the predictions is remarkable, achieving a high score quantified by the cumulative BLEU 4-gram, a common metric in language recognition studies.

Speaker verification aims to verify whether an input speech corresponds to the claimed speaker, and conventionally, this kind of system is deployed based on single-stream scenario, wherein the feature extractor operates in full frequency range. In this paper, we hypothesize that machine can learn enough knowledge to do classification task when listening to partial frequency range instead of full frequency range, which is so called frequency selection technique, and further propose a novel framework of multi-stream Convolutional Neural Network (CNN) with this technique for speaker verification tasks. The proposed framework accommodates diverse temporal embeddings generated from multiple streams to enhance the robustness of acoustic modeling. For the diversity of temporal embeddings, we consider feature augmentation with frequency selection, which is to manually segment the full-band of frequency into several sub-bands, and the feature extractor of each stream can select which sub-bands to use as target frequency domain. Different from conventional single-stream solution wherein each utterance would only be processed for one time, in this framework, there are multiple streams processing it in parallel. The input utterance for each stream is pre-processed by a frequency selector within specified frequency range, and post-processed by mean normalization. The normalized temporal embeddings of each stream will flow into a pooling layer to generate fused embeddings. We conduct extensive experiments on VoxCeleb dataset, and the experimental results demonstrate that multi-stream CNN significantly outperforms single-stream baseline with 20.53 % of relative improvement in minimum Decision Cost Function (minDCF).

Large Language Models (LLMs) inherently reflect the vast data distributions they encounter during their pre-training phase. As this data is predominantly sourced from the web, there is a high chance it will be skewed towards high-resourced languages and cultures, such as those of the West. Consequently, LLMs often exhibit a diminished understanding of certain communities, a gap that is particularly evident in their knowledge of Arabic and Islamic cultures. This issue becomes even more pronounced with increasingly under-represented topics. To address this critical challenge, we introduce PalmX 2025, the first shared task designed to benchmark the cultural competence of LLMs in these specific domains. The task is composed of two subtasks featuring multiple-choice questions (MCQs) in Modern Standard Arabic (MSA): General Arabic Culture and General Islamic Culture. These subtasks cover a wide range of topics, including traditions, food, history, religious practices, and language expressions from across 22 Arab countries. The initiative drew considerable interest, with 26 teams registering for Subtask 1 and 19 for Subtask 2, culminating in nine and six valid submissions, respectively. Our findings reveal that task-specific fine-tuning substantially boosts performance over baseline models. The top-performing systems achieved an accuracy of 72.15% on cultural questions and 84.22% on Islamic knowledge. Parameter-efficient fine-tuning emerged as the predominant and most effective approach among participants, while the utility of data augmentation was found to be domain-dependent.

Post-training of Large Language Models (LMs) often prioritizes accuracy and helpfulness at the expense of diversity. This creates a tension: while post-training improves response quality, it also sharpens output distributions and reduces the range of ideas, limiting the usefulness of LMs in creative and exploratory tasks such as brainstorming, storytelling, or problem solving. We address this challenge with Diversity-Aware Reinforcement Learning (DARLING), a framework that jointly optimizes for response quality and semantic diversity. At its core, DARLING introduces a learned partition function to measure diversity beyond surface-level lexical variations. This diversity signal is then combined with a quality reward during online reinforcement learning, encouraging models to generate outputs that are both high-quality and distinct. Experiments across multiple model families and sizes show that DARLING generalizes to two regimes: non-verifiable tasks (instruction following and creative writing) and verifiable tasks (competition math). On five benchmarks in the first setting, DARLING consistently outperforms quality-only RL baselines, producing outputs that are simultaneously of higher quality and novelty. In the second setting, DARLING achieves higher pass@1 (solution quality) and pass@k (solution variety). Most strikingly, explicitly optimizing for diversity catalyzes exploration in online RL, which manifests itself as higher-quality responses.

Recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) have empowered large language models (LLMs) to tackle challenging reasoning tasks such as mathematics and programming. RLVR leverages verifiable outcome rewards to guide policy optimization, enabling LLMs to progressively improve output quality in a grounded and reliable manner. Despite its promise, the RLVR paradigm poses significant challenges, as existing methods often suffer from sparse reward signals and unstable policy gradient updates, particularly in RL-based approaches. To address the challenges, we propose $\textbf{PACS}$, a novel RLVR framework that achieves im$\textbf{P}$licit $\textbf{A}$ctor $\textbf{C}$ritic coupling via a $\textbf{S}$upervised learning framework. By treating the outcome reward as a predictable label, we reformulate the RLVR problem into a supervised learning task over a score function parameterized by the policy model and optimized using cross-entropy loss. A detailed gradient analysis shows that this supervised formulation inherently recovers the classical policy gradient update while implicitly coupling actor and critic roles, yielding more stable and efficient training. Benchmarking on challenging mathematical reasoning tasks, PACS outperforms strong RLVR baselines, such as PPO and GRPO, achieving superior reasoning performance. For instance, PACS achieves 59.78\% at pass@256 on AIME 2025, representing improvements of 13.32 and 14.36 points over PPO and GRPO. This simple yet powerful framework offers a promising avenue for LLMs post-training with verifiable rewards. Our code and data are available as open source at https://github.com/ritzz-ai/PACS.