Multimodal learning has shown promise in medical imaging, combining complementary modalities like images and text. Vision-language models (VLMs) capture rich diagnostic cues but often require large paired datasets and promptor text-based inference. Their practicality is therefore limited due to annotation cost, privacy, and compute demands. Unpaired external text, like pathology reports, can still provide complementary diagnostic cues if semantically relevant content is retrievable per image. To address this, we introduce CLIP-IT, a novel framework that relies on rich unpaired text reports. Specifically, CLIP-IT uses a CLIP model pre-trained on histology image–text pairs from a separate dataset to retrieve the most relevant unpaired textual report for each image in the downstream unimodal dataset. These reports, sourced from the same disease domain and tissue type, form pseudo-pairs that reflect shared clinical semantics rather than exact alignment. Knowledge from these texts is distilled into the vision model during training, while LoRA-based adaptation mitigates the semantic gap between unaligned modalities. At inference, only the vision model is used, maintaining low overhead while still benefiting from multimodal training without requiring paired data in the downstream dataset. Experiments1show that CLIP-IT consistently improves classification accuracy over both unimodal and multimodal CLIP-based baselines in most cases, without requiring paired annotations per dataset or incurring additional inference-time complexity.

Casting complex inputs into tractable representations is a critical step across various fields. Diverse embedding models emerge from differences in architectures, loss functions, input modalities and datasets, each capturing unique aspects of the input. Multi-teacher distillation leverages this diversity to enrich representations but often remains tailored to specific tasks. We introduce a task-agnostic framework based on a ``majority vote" objective function. We demonstrate that this function is bounded by the mutual information between the student and the teachers' embeddings, leading to a task-agnostic distillation loss that eliminates dependence on task-specific labels or prior knowledge. Comprehensive evaluations across text, vision models, and molecular modeling show that our method effectively leverages teacher diversity, resulting in representations enabling better performance for a wide range of downstream tasks such as classification, clustering, or regression. Additionally, we train and release state-of-the-art embedding models, enhancing downstream performance in various modalities.

Classifier combination through ensemble systems is one of the most effective approaches to improve the accuracy of classification systems. Ensemble systems are generally used to combine classifiers; However, selecting the best combination of individual classifiers is a challenging task. In this paper, we propose an efficient assembling method that employs both meta-learning and a genetic algorithm for the selection of the best classifiers. Our method is called MEGA, standing for using MEta-learning and a Genetic Algorithm for algorithm recommendation. MEGA has three main components: Training, Model Interpretation and Testing. The Training component extracts meta-features of each training dataset and uses a genetic algorithm to discover the best classifier combination. The Model Interpretation component interprets the relationships between meta-features and classifiers using a priori and multi-label decision tree algorithms. Finally, the Testing component uses a weighted k-nearest-neighbors algorithm to predict the best combination of classifiers for unseen datasets. We present extensive experimental results that demonstrate the performance of MEGA. MEGA achieves superior results in a comparison of three other methods and, most importantly, is able to find novel interpretable rules that can be used to select the best combination of classifiers for an unseen dataset.