IDEAL: Interpretable-by-Design ALgorithms for learning from foundation feature spaces
Abstract: Many of the existing transfer learning methods rely on parametric tuning and lack interpretation of decision making. However, the advance of foundation models (FM) makes it possible to avoid such parametric tuning, taking advantage of pretrained feature spaces. In this study, we define a framework called IDEAL (Interpretable-by-design DEep learning ALgorithms) which tackles the problem of interpretable transfer learning by recasting the standard supervised classification problem into a function of similarity to a set of prototypes derived from the training data. This framework generalises previously-known prototypical approaches such as ProtoPNet, xDNN and DNC.} Using the IDEAL approach we can decompose the overall problem into two inherently connected stages: A) feature extraction (FE), which maps the raw features of the real-world data into a latent space, and B) identification of representative prototypes and decision making based on similarity and association between the query and the prototypes. This addresses the issue of interpretability (stage B) while retaining the benefits from the tremendous achievements offered by deep learning (DL) models (e.g., visual transformers, ViT) which are often pre-trained on huge datasets such as IG-3.6B + ImageNet-1K or LVD-142M (stage A).
On a range of datasets (CIFAR-10, CIFAR-100, CalTech101, STL-10, Oxford-IIIT Pet, EuroSAT), we demonstrate, through an extensive set of experiments, how the choice of the latent space, prototype selection, and finetuning of the latent space affect accuracy and generalisation of the models on transfer learning scenarios for different backbones. Building upon this knowledge, we demonstrate that the proposed framework helps achieve an advantage over state-of-the-art baselines in class-incremental learning. Finally, we analyse the interpretations provided by the proposed IDEAL framework, as well as the impact of confounding in transfer learning, demonstrating that the proposed approach without finetuning improves the performance on confounded data over finetuned counterparts.
The key findings can be summarized as follows: (1) the setting allows interpretability through prototypes, while also mitigating the issue of confounding bias, (2) lack of finetuning helps circumvent the issue of catastrophic forgetting, allowing efficient class-incremental transfer learning, and (3) ViT architectures narrow the gap between finetuned and non-finetuned models allowing for transfer learning in a fraction of time without finetuning of the feature space on a target dataset with iterative supervised methods.
Submission Length: Long submission (more than 12 pages of main content)
Assigned Action Editor: ~Kui_Jia1
Submission Number: 1879
Loading