Accurate disease interpretation from radiology remains challenging due to imaging heterogeneity. Achieving expert-level diagnostic decisions requires integration of subtle image features with clinical knowledge. Yet major vision-language models (VLMs) treat images as holistic entities and overlook fine-grained image details that are vital for disease diagnosis. Clinicians analyze images by utilizing their prior medical knowledge and identify anatomical structures as important region of interests (ROIs). Inspired from this human-centric workflow, we introduce Anatomy-VLM, a fine-grained, vision-language model that incorporates multi-scale information. First, we design a model encoder to localize key anatomical features from entire medical images. Second, these regions are enriched with structured knowledge for contextually-aware interpretation. Finally, the model encoder aligns multi-scale medical information to generate clinically-interpretable disease prediction. Anatomy-VLM achieves outstanding performance on both in- and out-of-distribution datasets. We also validate the performance of Anatomy-VLM on downstream image segmentation tasks, suggesting that its fine-grained alignment captures anatomical and pathology-related knowledge. Furthermore, the Anatomy-VLM's encoder facilitates zero-shot anatomy-wise interpretation, providing its strong expert-level clinical interpretation capabilities.

Among the many desirable properties of Artificial Intelligence systems, sustainability and efficiency have become increasingly important in the context of worsening climate change, massive water use in data centres, and the need for simpler, faster models in IoT settings. Consequently, there have been not few attempts to both promote and regulate the sustainability of Machine Learning models; the EU's AI Act indicates that the sustainability of AI - in terms of its environmental and social footprint-should be considered when developing and deploying AI pipelines [1], and manifests like that of UNESCO highlight sustainability as one of the core principles of the broader Responsible AI paradigm [2]. However, this seemingly consensual agreement on the importance of sustainability and efficiency for real-world AI systems and the social and regulatory efforts heavily contrasts with the practical applicability of such regulations; without looking further, the AI Act itself defines the requirement for sustainability, but does not indicate what metrics and evaluation pipelines should be considered for a robust, reliable, and practically relevant assessment of the environmental impact of a model. We argue that this lack of comprehensiveness in sustainability recommendations for AI systems does not stem from a careless or sloppy construction of the regulations themselves, but rather from an actual absence of suitable evaluation protocols that are formal, model-agnostic, reproducible, and grounded in real-life usage protocols for the ML lifecycle. The authors of this preprint are aware of a single regulatory standard for measuring AI sustainability, namely UNE 0086 [3], which limits evaluation to the epoch-batch training paradigm of supervised learning systems, rendering it useless for any task or type of learning that deviates from that standard. Although many researchers and companies have made it a habit to report efficiency figures and comparisons (e.g., in terms of emitted CO

Weakly supervised learning often operates with coarse aggregate signals rather than instance labels. We study a setting where each training example is an $n$-tuple containing exactly m positives, while only the count m per tuple is observed. This NTMP (N-tuple with M positives) supervision arises in, e.g., image classification with region proposals and multi-instance measurements. We show that tuple counts admit a trainable unbiased risk estimator (URE) by linking the tuple-generation process to latent instance marginals. Starting from fixed (n,m), we derive a closed-form URE and extend it to variable tuple sizes, variable counts, and their combination. Identification holds whenever the effective mixing rate is separated from the class prior. We establish generalization bounds via Rademacher complexity and prove statistical consistency with standard rates under mild regularity assumptions. To improve finite-sample stability, we introduce simple ReLU corrections to the URE that preserve asymptotic correctness. Across benchmarks converted to NTMP tasks, the approach consistently outperforms representative weak-supervision baselines and yields favorable precision-recall and F1 trade-offs. It remains robust under class-prior imbalance and across diverse tuple configurations, demonstrating that count-only supervision can be exploited effectively through a theoretically grounded and practically stable objective.

Transformers replace recurrence with a memory that grows with sequence length and self-attention that enables ad-hoc look ups over past tokens. Consequently, they lack an inherent incentive to compress history into compact latent states with consistent transition rules. This often leads to learning solutions that generalize poorly. We introduce Next-Latent Prediction (NextLat), which extends standard next-token training with self-supervised predictions in the latent space. Specifically, NextLat trains a transformer to learn latent representations that are predictive of its next latent state given the next output token. Theoretically, we show that these latents provably converge to belief states, compressed information of the history necessary to predict the future. This simple auxiliary objective also injects a recurrent inductive bias into transformers, while leaving their architecture, parallel training, and inference unchanged. NextLat effectively encourages the transformer to form compact internal world models with its own belief states and transition dynamics -- a crucial property absent in standard next-token prediction transformers. Empirically, across benchmarks targeting core sequence modeling competencies -- world modeling, reasoning, planning, and language modeling -- NextLat demonstrates significant gains over standard next-token training in downstream accuracy, representation compression, and lookahead planning. NextLat stands as a simple and efficient paradigm for shaping transformer representations toward stronger generalization.

Many real-world decisions are made under uncertainty by solving optimization problems using predicted quantities. This predict-then-optimize paradigm has motivated decision-focused learning, which trains models with awareness of how the optimizer uses predictions, improving the performance of downstream decisions. Despite its promise, scaling is challenging: state-of-the-art methods either differentiate through a solver or rely on task-specific surrogates, both of which require frequent and expensive calls to an optimizer, often a combinatorial one. In this paper, we leverage dual variables from the downstream problem to shape learning and introduce Dual-Guided Loss (DGL), a simple, scalable objective that preserves decision alignment while reducing solver dependence. We construct DGL specifically for combinatorial selection problems with natural one-of-many constraints, such as matching, knapsack, and shortest path. Our approach (a) decouples optimization from gradient updates by solving the downstream problem only periodically; (b) between refreshes, trains on dual-adjusted targets using simple differentiable surrogate losses; and (c) as refreshes become less frequent, drives training cost toward standard supervised learning while retaining strong decision alignment. We prove that DGL has asymptotically diminishing decision regret, analyze runtime complexity, and show on two problem classes that DGL matches or exceeds state-of-the-art DFL methods while using far fewer solver calls and substantially less training time. Code is available at https://github.com/

Abstract--Wearable EEG devices have emerged as a promising alternative to polysomnography (PSG). As affordable and scalable solutions, their widespread adoption results in the collection of massive volumes of unlabeled data that cannot be analyzed by clinicians at scale. Meanwhile, the recent success of deep learning for sleep scoring has relied on large annotated datasets. Self-supervised learning (SSL) offers an opportunity to bridge this gap, leveraging unlabeled signals to address label scarcity and reduce annotation effort. In this paper, we present the first systematic evaluation of SSL for sleep staging using wearable EEG. We investigate a range of well-established SSL methods and evaluate them on two sleep databases acquired with the Ikon Sleep wearable EEG headband: BOAS, a high-quality benchmark containing PSG and wearable EEG recordings with consensus labels, and HOGAR, a large collection of home-based, self-recorded, and unlabeled recordings. Three evaluation scenarios are defined to study label efficiency, representation quality, and cross-dataset generalization. Results show that SSL consistently improves classification performance by up to 10% over supervised baselines, with gains particularly evident when labeled data is scarce. SSL achieves clinical-grade accuracy above 80% leveraging only 5% to 10% of labeled data, while the supervised approach requires twice the labels. Additionally, SSL representations prove robust to variations in population characteristics, recording environments, and signal quality . Our findings demonstrate the potential of SSL to enable label-efficient sleep staging with wearable EEG, reducing reliance on manual annotations and advancing the development of affordable sleep monitoring systems.

RDF pattern-based extraction is a compelling approach for fine-tuning small language models (SLMs) by focusing a relation extraction task on a specified SHACL shape. This technique enables the development of efficient models trained on limited text and RDF data. In this article, we introduce Kastor, a framework that advances this approach to meet the demands for completing and refining knowledge bases in specialized domains. Kastor reformulates the traditional validation task, shifting from single SHACL shape validation to evaluating all possible combinations of properties derived from the shape. By selecting the optimal combination for each training example, the framework significantly enhances model generalization and performance. Additionally, Kastor employs an iterative learning process to refine noisy knowledge bases, enabling the creation of robust models capable of uncovering new, relevant facts.

Recently, self-supervised contrastive learning has achieved great success on various tasks. However, its underlying working mechanism is yet unclear. In this paper, we first provide the tightest bounds based on the widely adopted assumption of conditional independence. Further, we relax the conditional independence assumption to a more practical assumption of augmentation overlap and derive the asymptotically closed bounds for the downstream performance. Our proposed augmentation overlap theory hinges on the insight that the support of different intra-class samples will become more overlapped under aggressive data augmentations, thus simply aligning the positive samples (augmented views of the same sample) could make contrastive learning cluster intra-class samples together. Moreover, from the newly derived augmentation overlap perspective, we develop an unsupervised metric for the representation evaluation of contrastive learning, which aligns well with the downstream performance almost without relying on additional modules. Code is available at https://github.com/PKU-ML/GARC.

Vision-Language Pretraining (VLP) has achieved remarkable success across various downstream tasks, but such gains are largely driven by scaling up on training data. Y et, literature methods treat image-text pairs as isolated training examples; this neglects the rich relational structure naturally present in many domains, such as e-commerce product co-purchase graphs and social recommendation networks. Inspired by neuroscientific evidence that human encodes knowledge as relationship cognitive maps, we introduce Structure-aware Language-Image Pretraining (SLIP). SLIP integrates a structural contrastive loss to align modalities while also modeling relationships between neighboring entities in a structured graph. To support this paradigm, we construct a large-scale Amazon Product Co-purchase Multi-modal Graph Dataset, enabling structured cross-modality supervision at scale. Experiment results show that SLIP consistently outperforms CLIP on cross-modal retrieval and classification tasks in both zero-shot and few-shot settings, showing the value of relational supervision for cross-modal alignment. Vision-language alignment has emerged as a key challenge in multimodal representation learning, with recent pretraining approaches achieving remarkable success by learning from web-scale data, driving progress in multimodal tasks such as image-text retrieval, visual question answering (VQA), and image captioning Gan et al. (2022). Ground-breaking work CLIP (Radford et al., 2021) has shown that a simple contrastive objective can yield state-of-the-art representations when scaled to millions of noisy image-text pairs, and such large-scale training has thus become the paradigm for vision-language foundation models. However, these web-scale corpora are notoriously noisy: captions can be generic, off-topic, or mismatched to the image.

The number of studies that combine Evolutionary Machine Learning and self-supervised learning has been growing steadily in recent years. Evolutionary Machine Learning has been shown to help automate the design of machine learning algorithms and to lead to more reliable solutions. Self-supervised learning, on the other hand, has produced good results in learning useful features when labelled data is limited. This suggests that the combination of these two areas can help both in shaping evolutionary processes and in automating the design of deep neural networks, while also reducing the need for labelled data. Still, there are no detailed reviews that explain how Evolutionary Machine Learning and self-supervised learning can be used together. To help with this, we provide an overview of studies that bring these areas together. Based on this growing interest and the range of existing works, we suggest a new sub-area of research, which we call Evolutionary Self-Supervised Learning and introduce a taxonomy for it. Finally, we point out some of the main challenges and suggest directions for future research to help Evolutionary Self-Supervised Learning grow and mature as a field.