Robust Contrastive Learning Using Negative Samples with Diminished Semantics
Dec 6, 2021
Speakers
Songwei Ge
Sprecher:in · 0 Follower:innen
Shlok Mishra
Sprecher:in · 0 Follower:innen
Haohan Wang
Sprecher:in · 0 Follower:innen
Chun-Liang Li
Sprecher:in · 1 Follower:in
David Jacobs
Sprecher:in · 0 Follower:innen
About
Unsupervised learning has recently made exceptional progress because of the development of more effective contrastive learning methods. However, CNNs are prone to depend on low-level features that humans deem non-semantic. This dependency has been conjectured to induce a lack of robustness to image perturbations or domain shift. In this paper, we show that by generating carefully designed negative samples, contrastive learning can learn more robust representations with less dependence on such features. Contrastive learning utilizes positive pairs which preserve semantic information while perturbing superficial features in the training images. Similarly, we propose to generate negative samples in a reversed way, where only the superfluous instead of the semantic features are preserved. We develop two methods, texture-based and patch-based augmentations, to generate negative samples. These samples achieve better generalization, especially under out-of-domain settings. We also analyze our method and the generated texture-based samples, showing that texture features are indispensable in classifying particular ImageNet classes and especially finer classes. We also show that the model bias between texture and shape features favors them differently under different test settings.Unsupervised learning has recently made exceptional progress because of the development of more effective contrastive learning methods. However, CNNs are prone to depend on low-level features that humans deem non-semantic. This dependency has been conjectured to induce a lack of robustness to image perturbations or domain shift. In this paper, we show that by generating carefully designed negative samples, contrastive learning can learn more robust representations with less dependence on such fe…
Organizer
NeurIPS 2021
Konto · 1,9k Follower:innen
About NeurIPS 2021
Neural Information Processing Systems (NeurIPS) is a multi-track machine learning and computational neuroscience conference that includes invited talks, demonstrations, symposia and oral and poster presentations of refereed papers. Following the conference, there are workshops which provide a less formal setting.
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Width-based Lookaheads with Learnt Base Policies and Heuristics Over the Atari-2600 Benchmark
Später ansehen
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Strength Through Diversity: Robust Behavior Learning via Mixture Policies
Später ansehen
Tim Seyde, …
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Physics-Based Learned Diffuser for Single-shot 3D Imaging
Später ansehen
Eric Markley, …
Ewigspeicher-Fortschrittswert: 0 = 0.0%
A Modern Self-Referential Weight Matrix That Learns to Modify Itself
Später ansehen
Kazuki Irie, …
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Joint Inference and input Optimization with Deep Equilibrium Models
Später ansehen
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Actively Identifying Causal Effects with Latent Variables Given Only Response Variable Observable
Später ansehen
Ewigspeicher-Fortschrittswert: 0 = 0.0%