{
  "id": "2503.07851",
  "version": "v2",
  "published": "2025-03-10T20:56:54.000Z",
  "updated": "2025-05-16T13:58:49.000Z",
  "title": "TwinTURBO: Semi-Supervised Fine-Tuning of Foundation Models via Mutual Information Decompositions for Downstream Task and Latent Spaces",
  "authors": [
    "Guillaume Quétant",
    "Pavlo Molchanov",
    "Slava Voloshynovskiy"
  ],
  "categories": [
    "cs.LG",
    "cs.CV",
    "cs.IT",
    "math.IT",
    "stat.ML"
  ],
  "abstract": "We present a semi-supervised fine-tuning framework for foundation models that utilises mutual information decomposition to address the challenges of training for a limited amount of labelled data. Our approach derives two distinct lower bounds: i) for the downstream task space, such as classification, optimised using conditional and marginal cross-entropy alongside Kullback-Leibler divergence, and ii) for the latent space representation, regularised and aligned using a contrastive-like decomposition. This fine-tuning strategy retains the pre-trained structure of the foundation model, modifying only a specialised projector module comprising a small transformer and a token aggregation technique. Experiments on several datasets demonstrate significant improvements in classification tasks under extremely low-labelled conditions by effectively leveraging unlabelled data.",
  "revisions": [
    {
      "version": "v2",
      "updated": "2025-05-16T13:58:49.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "foundation model",
      "downstream task",
      "latent space",
      "semi-supervised fine-tuning",
      "marginal cross-entropy alongside kullback-leibler divergence"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}