Guarantees for tuning step size using meta-gradient descent.
problem Choosing optimal parameters for optimization algorithms.
method Learning-to-learn approach with meta-gradient descent.
result Meta-gradient can explode/vanish, requiring careful meta-objective design.
Meta-gradient RL learns to learn from experience.
problem Learning efficiency and adaptability in reinforcement learning.
method Meta-gradient descent to discover its own objective function.
result Meta-gradient RL adapts to learn more efficiently over time.
Meta-gradient D4PG optimizes performance and constraint adherence in RL.
problem Balancing performance and adherence to complex constraints in RL.
method Uses meta-gradients to find a balance between expected return and minimizing constraint violations.
result Meta-gradient D4PG consistently outperforms baselines across MuJoCo domains.
Evolutionary Strategies optimize hyper-parameters for off-policy learning.
problem Hyper-parameter sensitivity in off-policy learning.
method Application of Evolutionary Strategies for online hyper-parameter tuning.
result Our method outperforms state-of-the-art baselines.
Proposes a meta-learning algorithm to improve semi-supervised learning.
problem Overfitting and limited model generalization in consistency-based semi-supervised learning.
method Introduces a learn-to-generalize regularization term and meta-gradient optimization.
result Demonstrates improved performance on various datasets compared to state-of-the-art methods.
Synthetic data helps prevent forgetting when learning sequentially.
problem Catastrophic forgetting in neural networks.
method Generate synthetic data via two-step optimisation process using meta-gradients.
result Training on synthetic data prevents forgetting when learning sequentially.
Accelerates policy optimization in RL with optimistic and adaptive updates.
problem Improving policy optimization methods in reinforcement learning.
method Integrates foresight into policy improvement step via optimistic and adaptive updates.
result Designs an optimistic policy gradient algorithm, adaptive via meta-gradient learning.
Representations are fundamental to artificial intelligence. The performance of a learning system depends on the type of representation used for representing the data. Typically, these representations are hand-engineered using domain knowledge. More recently, the trend is to learn these representations through stochasti…
AdvImmune improves certifiable robustness of GNNs against adversarial attacks.
problem Vulnerability of graph neural networks to adversarial attacks.
method Proposes AdvImmune, an algorithm that optimizes with meta-gradient to improve certifiable robustness.
result Remarkably improves the ratio of robust nodes by 12%, 42%, 65% with an affordable immune budget of only 5% edges.
A core capability of intelligent systems is the ability to quickly learn new tasks by drawing on prior experience. Gradient (or optimization) based meta-learning has recently emerged as an effective approach for few-shot learning. In this formulation, meta-parameters are learned in the outer loop, while task-specific m…
Reinforcement learning (RL) has had many successes in both "deep" and "shallow" settings. In both cases, significant hyperparameter tuning is often required to achieve good performance. Furthermore, when nonlinear function approximation is used, non-stationarity in the state representation can lead to learning instabil…
Deep learning models for graphs have advanced the state of the art on many tasks. Despite their recent success, little is known about their robustness. We investigate training time attacks on graph neural networks for node classification that perturb the discrete graph structure. Our core principle is to use meta-gradi…
Improves meta-learning efficiency with mixed-mode differentiation.
problem Efficiently calculating complex derivatives in meta-learning.
method Mixed-Flow Meta-Gradients (MixFlow-MG) for scalable differentiation.
result Significant memory and time improvements in meta-learning tasks.
DataRater learns which data points are most valuable for training models.
problem Training model efficiency depends on high-quality training data.
method Meta-learning to estimate the value of data points for training.
result Meta-learning improves compute efficiency by filtering data effectively.
VIABLE learns a loss function for better few-shot learning.
problem Few-shot learning underfits with standard loss functions.
method Meta-learning to learn a differentiable loss function.
result Learning a relational loss function improves performance and sample efficiency.
The goal of reinforcement learning algorithms is to estimate and/or optimise the value function. However, unlike supervised learning, no teacher or oracle is available to provide the true value function. Instead, the majority of reinforcement learning algorithms estimate and/or optimise a proxy for the value function. …
MetAL improves graph classification models with fewer labeled data.
problem Efficiently selecting unlabeled graph instances for training.
method Formulates AL as bilevel optimization, uses meta-learning to approximate model performance.
result MetAL outperforms existing AL algorithms on multiple graph datasets.
New algorithm improves dataset distillation with 108% improvement on ImageNet.
problem Improving dataset distillation performance.
method Reparameterization and convexification of implicit gradients (RCIG).
result Establishes new state-of-the-art on various dataset distillation tasks.
MSLG generates soft labels to improve DNN performance on noisy datasets.
problem Significant performance degradation of DNNs due to noisy labels.
method Meta-learning techniques to estimate optimal label distribution and iteratively update soft labels.
result MSLG outperforms state-of-the-art methods by a large margin on various datasets.
Deep neural networks have been shown to be very powerful modeling tools for many supervised learning tasks involving complex input patterns. However, they can also easily overfit to training set biases and label noises. In addition to various regularizers, example reweighting algorithms are popular solutions to these p…
This paper investigates different vector step-size adaptation approaches for non-stationary online, continual prediction problems. Vanilla stochastic gradient descent can be considerably improved by scaling the update with a vector of appropriately chosen step-sizes. Many methods, including AdaGrad, RMSProp, and AMSGra…
New method optimises learning via surrogate PAC-Bayes bounds.
problem Computational intractability of optimising generalisation bounds.
method Iteratively optimising surrogate training objectives derived from PAC-Bayes bounds.
result Iteratively optimising surrogates implies optimising original generalisation bounds.
New framework guarantees convergence of multi-step MAML.
problem Convergence of multi-step MAML in nonconvex settings.
method Developed a theoretical framework for two types of MAML objective functions.
result Guaranteed convergence rate and computational complexity for multi-step MAML.
Self-Tuning Actor-Critic improves reinforcement learning performance.
problem Manual hyperparameter tuning is time-consuming and domain-specific.
method Uses metagradients for online hyperparameter adaptation.
result Improves performance across various domains and tasks.
Improved learning to reweight using deep interactions between student and teacher models.
problem Limitation of existing learning to reweight methods in utilizing student model's internal states.
method Proposes an algorithm that uses the student model's internal states to the teacher model, which returns adaptive weights to enhance student model training.
result Significant improvement over previous methods in image classification and neural machine translation experiments.
G-Meta learns graph meta-learning from local subgraphs.
problem Learning from scarce data in graph tasks.
method Uses local subgraphs to transfer subgraph-specific information and learn transferable knowledge.
result G-Meta outperforms existing methods by up to 16.3% on seven datasets.
Learning an efficient update rule from data that promotes rapid learning of new tasks from the same distribution remains an open problem in meta-learning. Typically, previous works have approached this issue either by attempting to train a neural network that directly produces updates or by attempting to learn better i…
GTNs generate training data to accelerate AI learning.
problem Speeding up AI learning through better training data.
method Generative Teaching Networks (GTNs) learn to generate synthetic training data.
result GTNs accelerate AI learning and NAS evaluations.