Understanding the Limits of Unsupervised Domain Adaptation via Data Poisoning
Dec 6, 2021
Speakers
Akshay Mehra
Speaker · 0 followers
Jihun Hamm
Speaker · 0 followers
Bhavya Kailkhura
Speaker · 0 followers
Pin-Yu Chen
Speaker · 3 followers
About
Unsupervised domain adaptation (UDA) enables cross-domain learning without target domain labels by transferring knowledge from a labeled source domain whose distribution differs from the target. However, UDA is not always successful and several accounts of `negative transfer' have been reported in the literature. In this work, we prove a simple lower bound on the target domain error that complements the existing upper bound. The bound shows the insufficiency of minimizing source domain error and marginal distribution mismatch for a guaranteed reduction in the target domain error, due to the possible increase of induced labeling function mismatch. This insufficiency is further illustrated through simple distributions for which the same UDA approach succeeds, fail, and may succeed or fail with an equal chance. Motivated from this, we propose novel data poisoning attacks to fool UDA methods into learning representations that produce large target domain error. We evaluate the effect of these attacks on popular UDA methods using benchmark datasets where they have been previously shown to be successful. Our results show that poisoning can significantly decrease the target domain accuracy, dropping it to almost 0% in some cases, with the addition of only 10% poisoned data in the source domain. The failure of UDA methods demonstrates the limitations of UDA at guaranteeing cross-domain generalization consistent with the lower bound. Thus, evaluation of UDA methods in adversarial settings such as data poisoning can provide a better sense of their robustness in scenarios unfavorable for UDA.Unsupervised domain adaptation (UDA) enables cross-domain learning without target domain labels by transferring knowledge from a labeled source domain whose distribution differs from the target. However, UDA is not always successful and several accounts of `negative transfer' have been reported in the literature. In this work, we prove a simple lower bound on the target domain error that complements the existing upper bound. The bound shows the insufficiency of minimizing source domain error and…
Organizer
NeurIPS 2021
Account · 1.9k followers
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
Parametrized Quantum Policies for Reinforcement Learning
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Robust Visual Reasoning via Language Guided Neural Module Networks
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
E(2) Equivariant Self-Attention for Radio Astronomy: Imperfect models and where to find them
Watch later
Micah Bowles, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Rates of Estimation of Optimal Transport Maps using Plug-in Estimators via Barycentric Projection
Watch later
Nabarun Deb, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Passive Attention in Artificial Neural Networks Predicts Human Visual Selectivity
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Concentration Inequalities Under Sub-Gaussian and Sub-Exponential Conditions
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%