# Task-Robust Model-Agnostic Meta-Learning Source: [Task-Robust Model-Agnostic Meta-Learning]() Collection index: [llms.txt]() Data updated: 2025-10-25T14:29:49.710Z Authors: Liam Collins, Aryan Mokhtari, Sanjay Shakkottai Published: 2020-02-12T02:20:51.000Z ## Abstract Meta-learning methods have shown an impressive ability to train models that rapidly learn new tasks. However, these methods only aim to perform well in expectation over tasks coming from some particular distribution that is typically equivalent across meta-training and meta-testing, rather than considering worst-case task performance. In this work we introduce the notion of "task-robustness" by reformulating the popular Model-Agnostic Meta-Learning (MAML) objective \[Finn et al. 2017\] such that the goal is to minimize the maximum loss over the observed meta-training tasks. The solution to this novel formulation is task-robust in the sense that it places equal importance on even the most difficult and/or rare tasks. This also means that it performs well over all distributions of the observed tasks, making it robust to shifts in the task distribution between meta-training and meta-testing. We present an algorithm to solve the proposed min-max problem, and show that it converges to an $\\epsilon$-accurate point at the optimal rate of $\\mathcal{O}(1/\\epsilon^2)$ in the convex setting and to an $(\\epsilon, \\delta)$-stationary point at the rate of $\\mathcal{O}(\\max\\{1/\\epsilon^5, 1/\\delta^5\\})$ in nonconvex settings. We also provide an upper bound on the new task generalization error that captures the advantage of minimizing the worst-case task loss, and demonstrate this advantage in sinusoid regression and image classification experiments. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Few-shot Image Classification on mini-ImageNet (Four Meta-Testing Tasks)](): Few-shot Image Classification; mini-ImageNet (Four Meta-Testing Tasks); Average Accuracy: 48.5. - [Few-shot image classification on Omniglot Alphabets (meta-test)](): Few-shot Image Classification; Omniglot Alphabets (meta-test); Average Score: 93.1. - [Few-shot Image Classification on mini-ImageNet Eight Meta (train)](): Few-shot Image Classification; mini-ImageNet Eight Meta (train); Average Accuracy: 63.2. - [Few-shot Sinusoid Regression on Sinusoid 490 tasks (5-shot, test)](): Few-shot Sinusoid Regression; Sinusoid 490 tasks 5-shot (test); Avg MSE: 1.09. - [Few-shot Sinusoid Regression on Sinusoid 490 meta-test tasks 10-shot (test)](): Few-shot Sinusoid Regression; Sinusoid 490 meta-test tasks 10-shot (test); Average MSE: 0.77. - [Few-shot image classification on Omniglot Meta-Training Alphabets (train)](): Few-shot Image Classification; Omniglot Meta-Training Alphabets (train); Average Performance: 97.4. - [Image Classification on mini-ImageNet (train)](): Image Classification; mini-ImageNet (train); Average Score: 62.2. - [System Identification on Pendulum (test)](): System Identification; Pendulum (test); Average MSE: 0.76. ## Original sources and implementation [Original paper]() [PDF]()