The application of Vision-Language Models (VLMs) in medicine is critically hampered by the scarcity of high-quality, expert-annotated data. Supervised Fine-Tuning (SFT) on existing datasets often leads to poor generalization on unseen modalities and tasks, while Reinforcement Learning (RL), a promising alternative, is stymied by the lack of reliable reward signals in this data-scarce domain. To break this impasse, we introduce Generative Reward Learning for Medical Reasoning (MedGR$^2$), a novel framework that creates a self-improving virtuous cycle. MedGR$^2$ co-develops a data generator and a reward model, enabling the automated, continuous creation of high-quality, multi-modal medical data that serves as both a superior training source for SFT and RL. Our experiments demonstrate that SFT with MedGR$^2$-produced data already surpasses baselines trained on large-scale, human-curated datasets. Crucially, when leveraging this data for RL via Group Relative Policy Optimization (GRPO), our model achieves state-of-the-art cross-modality and cross-task generalization, significantly outperforming specialized RL-based methods. Furthermore, our compact model, empowered by MedGR$^2$, achieves performance competitive with foundation models possessing over 10 times more parameters. MedGR$^2$ presents a new paradigm for data-efficient learning in high-stakes domains, transforming the problem from data scarcity to data generation and unlocking the full potential of RL for building truly generalizable medical AI.

Magnetic resonance imaging (MRI) inpainting supports numerous clinical and research applications. We introduce the first generative model that conditions on voxel-level, continuous tumor concentrations to synthesize high-fidelity brain tumor MRIs. For the BraTS 2025 Inpainting Challenge, we adapt this architecture to the complementary task of healthy tissue restoration by setting the tumor concentrations to zero. Our latent diffusion model conditioned on both tissue segmentations and the tumor concentrations generates 3D spatially coherent and anatomically consistent images for both tumor synthesis and healthy tissue inpainting. For healthy inpainting, we achieve a PSNR of 18.5, and for tumor inpainting, we achieve 17.4. Our code is available at: https://github.com/valentin-biller/ldm.git

Significant Others (SOs) stabilize identity, regulate emotion, and support narrative meaning-making, yet many people today lack access to such relational anchors. Recent advances in large language models and memory-augmented AI raise the question of whether artificial systems could support some of these functions. Existing empathic AIs, however, remain reactive and short-term, lacking autobiographical memory, identity modeling, predictive emotional regulation, and narrative coherence. This manuscript introduces Significant Other Artificial Intelligence (SO-AI) as a new domain of relational AI. It synthesizes psychological and sociological theory to define SO functions and derives requirements for SO-AI, including identity awareness, long-term memory, proactive support, narrative co-construction, and ethical boundary enforcement. A conceptual architecture is proposed, comprising an anthropomorphic interface, a relational cognition layer, and a governance layer. A research agenda outlines methods for evaluating identity stability, longitudinal interaction patterns, narrative development, and sociocultural impact. SO-AI reframes AI-human relationships as long-term, identity-bearing partnerships and provides a foundational blueprint for investigating whether AI can responsibly augment the relational stability many individuals lack today.

Large Language Models (LLMs) offer powerful capabilities, but their significant size and computational requirements hinder deployment on resource-constrained mobile devices. This paper investigates Post-Training Quantization (PTQ) for compressing LLMs for mobile execution. We apply 4-bit PTQ using the BitsAndBytes library with the Hugging Face Transformers framework to Meta's Llama 3.2 3B model. The quantized model is converted to GGUF format using llama.cpp tools for optimized mobile inference. The PTQ workflow achieves a 68.66% reduction in model size through 4-bit quantization, enabling the Llama 3.2 3B model to run efficiently on an Android device. Qualitative validation shows that the 4-bit quantized model can perform inference tasks successfully. We demonstrate the feasibility of running the quantized GGUF model on an Android device using the Termux environment and the Ollama framework. PTQ, especially at 4-bit precision combined with mobile-optimized formats like GGUF, provides a practical pathway for deploying capable LLMs on mobile devices, balancing model size and performance.

Assessing language proficiency is essential for education, as it enables instruction tailored to learners needs. This paper investigates the use of Large Language Models (LLMs) for automatically classifying German texts according to the Common European Framework of Reference for Languages (CEFR) into different proficiency levels. To support robust training and evaluation, we construct a diverse dataset by combining multiple existing CEFR-annotated corpora with synthetic data. We then evaluate prompt-engineering strategies, fine-tuning of a LLaMA-3-8B-Instruct model and a probing-based approach that utilizes the internal neural state of the LLM for classification. Our results show a consistent performance improvement over prior methods, highlighting the potential of LLMs for reliable and scalable CEFR classification.

Autonomous driving (AD) scenarios are inherently complex and diverse, posing significant challenges for a single deep learning model to effectively cover all possible conditions, such as varying weather, traffic densities, and road types. Large Model (LM)-Driven Mixture of Experts (MoE) paradigm offers a promising solution, where LM serves as the backbone to extract latent features while MoE serves as the downstream head to dynamically select and aggregate specialized experts to adapt to different scenarios. However, routing and aggregating in MoE face intrinsic challenges, including imprecise expert selection due to flawed routing strategy and inefficient expert aggregation leading to suboptimal prediction. To address these issues, we propose a statistic-augmented, decoupled MoE }outing and Aggregating Mechanism (MoE-RAM) driven by LM. Specifically, on the one hand, MoE-RAM enhances expert routing by incorporating statistical retrieval mechanism to match LM-extracted latent features with cached prototypical features of the most relevant experts; on the other hand, MoE-RAM adaptively reweights experts' outputs in fusion by measuring statistical distances of experts' instant features against LM-extracted latent features. Benefiting from the synergy of the statistic-augmented MoE's routing and aggregating, MoE-RAM ultimately improves the prediction performance. We take the AD semantic segmentation task as an example to assess the proposed MoE-RAM. Extensive experiments on AD datasets demonstrate the superiority of MoE-RAM compared to other MoE baselines and conventional single-model approaches.

Analysts in Security Operations Centers routinely query massive telemetry streams using Kusto Query Language (KQL). Writing correct KQL requires specialized expertise, and this dependency creates a bottleneck as security teams scale. This paper investigates whether Small Language Models (SLMs) can enable accurate, cost-effective natural-language-to-KQL translation for enterprise security. We propose a three-knob framework targeting prompting, fine-tuning, and architecture design. First, we adapt existing NL2KQL framework for SLMs with lightweight retrieval and introduce error-aware prompting that addresses common parser failures without increasing token count. Second, we apply LoRA fine-tuning with rationale distillation, augmenting each NLQ-KQL pair with a brief chain-of-thought explanation to transfer reasoning from a teacher model while keeping the SLM compact. Third, we propose a two-stage architecture that uses an SLM for candidate generation and a low-cost LLM judge for schema-aware refinement and selection. We evaluate nine models (five SLMs and four LLMs) across syntax correctness, semantic accuracy, table selection, and filter precision, alongside latency and token cost. On Microsoft's NL2KQL Defender Evaluation dataset, our two-stage approach achieves 0.987 syntax and 0.906 semantic accuracy. We further demonstrate generalizability on Microsoft Sentinel data, reaching 0.964 syntax and 0.831 semantic accuracy. These results come at up to 10x lower token cost than GPT-5, establishing SLMs as a practical, scalable foundation for natural-language querying in security operations.

Understanding the underlying linguistic rules of plant genomes remains a fundamental challenge in computational biology. Recent advances including AgroNT and PDLLMs have made notable progress although, they suffer from excessive parameter size and limited ability to model the bidirectional nature of DNA strands respectively. To address these limitations, we propose PlantBiMoE, a lightweight and expressive plant genome language model that integrates bidirectional Mamba and a Sparse Mixture-of-Experts (SparseMoE) framework. The bidirectional Mamba enables the model to effectively capture structural dependencies across both the forward and reverse DNA strands, while SparseMoE significantly reduces the number of active parameters, improving computational efficiency without sacrificing modeling capacity. We evaluated and tested our model on the Modified Plants Genome Benchmark (MPGB), an enhanced genomic benchmark, which consolidates 31 datasets across 11 representative tasks, with input sequence lengths ranging from 50 to 6,000 bp. Experimental results demonstrate that PlantBiMoE achieves the best performance on 20 out of 31 datasets and the average best when comparing with existing models. In summary, all above results demonstrate that our model can effectively represent plant genomic sequences, serving as a robust computational tool for diverse genomic tasks, while making substantive contributions to plant genomics, gene editing, and synthetic biology. The code is available at: https://github.com/HUST-Keep-Lin/PlantBiMoE

Responding to Hodel et al.'s (2024) call for a formal definition of task relatedness in re-arc, we present the first 9-category taxonomy of all 400 tasks, validated at 97.5% accuracy via rule-based code analysis. We prove the taxonomy's visual coherence by training a CNN on raw grid pixels (95.24% accuracy on S3, 36.25% overall, 3.3x chance), then apply the taxonomy diagnostically to the original ARC-AGI-2 test set. Our curriculum analysis reveals 35.3% of tasks exhibit low neural affinity for Transformers--a distributional bias mirroring ARC-AGI-2. To probe this misalignment, we fine-tuned a 1.7M-parameter Transformer across 302 tasks, revealing a profound Compositional Gap: 210 of 302 tasks (69.5%) achieve >80% cell accuracy (local patterns) but <10% grid accuracy (global synthesis). This provides direct evidence for a Neural Affinity Ceiling Effect, where performance is bounded by architectural suitability, not curriculum. Applying our framework to Li et al.'s independent ViTARC study (400 specialists, 1M examples each) confirms its predictive power: Very Low affinity tasks achieve 51.9% versus 77.7% for High affinity (p<0.001), with a task at 0% despite massive data. The taxonomy enables precise diagnosis: low-affinity tasks (A2) hit hard ceilings, while high-affinity tasks (C1) reach 99.8%. These findings indicate that progress requires hybrid architectures with affinity-aligned modules. We release our validated taxonomy,

This Inaugural Lecture was given at Royal Holloway University of London in 1996. It covers an introduction to machine learning and describes various theoretical advances and practical projects in the field. The Lecture here is presented in its original format, but a few remarks have been added in 2025 to reflect recent developments, and the list of references has been updated to enhance the convenience and accuracy for readers. When did machine learning start? Maybe a good starting point is 1949, when Claude Shannon proposed a learning algorithm for chess-playing programs. Or maybe we should go back to the 1930s when Ronald Fisher developed discriminant analysis - a type of learning where the problem is to construct a decision rule that separates two types of vectors. Or could it be the 18th century when David Hume discussed the idea of induction? Or the 14th century, when William of Ockham formulated the principle of "simplicity" known as "Ockham's razor" (Ockham, by the way, is a small village not far from Royal Holloway). Or it may be that, like almost everything else in Western civilisation and culture, the origin of these ideas lies in the Mediterranean. After all, it was Aristotle who said that "we learn some things only by doing things". The field of machine learning has been greatly influenced by other disciplines and the subject is in itself not a very homogeneous discipline, but includes separate, overlapping subfields. There are many parallel lines of research in ML: inductive learning, neural networks, clustering, and theories of learning. They are all part of the more general field of machine learning.