Deep Learning
Compositional Generalization in Unsupervised Compositional Representation Learning: A Study on Disentanglement and Emergent Language
In this work, we evaluate these unsupervised learning algorithms in terms of how well they enable compositional generalization .
Robust and Decomposable Average Precision for Image Retrieval Elias Ramzi
In image retrieval, standard evaluation metrics rely on score ranking, e.g.
Appendix to TreeMoCo: Contrastive Neuron Morphology Representation Learning
The formulas of these two metrics are detailed in Sec. Figure A1 (a), by only re-ordering of edge vectors, a branch's shape can be significantly changed Such measurement is taken in a bag-of-word (BOW) fashion. The context and spatial relationship between arbors are ignored under this measurement. Branch A and B are composed of the same set of edge vectors. Each neuron is viewed from three different angles. Three public datasets are used in this study.
TreeMoCo: Contrastive Neuron Morphology Representation Learning
Traditional analysis mostly relies on heuristic features and visual inspections.
On the Loss Landscape of Adversarial Training: Identifying Challenges and How to Overcome Them Chen Liu
To this end, we first provide analytical studies of the properties of adversarial loss functions under different adversarial budgets. We then demonstrate that the adversarial loss landscape is less favorable to optimization, due to increased curvature and more scattered gradients.
Visual Prompting via Image Inpainting Amir Bar
The growing capacity of modern deep learning models made them prone to overfitting when trained on relatively small labeled datasets.
Learning Semantic Representations to Verify Hardware Designs
Practical verification relies on a "best effort" process that simulates the design on test inputs.