Modern meta-reinforcement learning (Meta-RL) methods are mainly developed based on model-agnostic meta-learning, which performs policy gradient steps across tasks to maximize policy performance. However, the gradient conflict problem is still poorly understood in Meta-RL, which may lead to performance degradation when encountering distinct tasks. To tackle this challenge, this paper proposes a novel personalized Meta-RL (pMeta-RL) algorithm, which aggregates task-specific personalized policies to update a meta-policy used for all tasks, while maintaining personalized policies to maximize the average return of each task under the constraint of the meta-policy. We also provide the theoretical analysis under the tabular setting, which demonstrates the convergence of our pMeta-RL algorithm. Moreover, we extend the proposed pMeta-RL algorithm to a deep network version based on soft actor-critic, making it suitable for continuous control tasks. Experiment results show that the proposed algorithms outperform other previous Meta-RL algorithms on Gym and MuJoCo suites.
Optimization-based meta-learning aims to learn an initialization so that a new unseen task can be learned within a few gradient updates. Model Agnostic Meta-Learning (MAML) is a benchmark algorithm comprising two optimization loops. The inner loop is dedicated to learning a new task and the outer loop leads to meta-initialization. However, ANIL (almost no inner loop) algorithm shows that feature reuse is an alternative to rapid learning in MAML. Thus, the meta-initialization phase makes MAML primed for feature reuse and obviates the need for rapid learning. Contrary to ANIL, we hypothesize that there may be a need to learn new features during meta-testing. A new unseen task from non-similar distribution would necessitate rapid learning in addition reuse and recombination of existing features. In this paper, we invoke the width-depth duality of neural networks, wherein, we increase the width of the network by adding extra computational units (ACU). The ACUs enable the learning of new atomic features in the meta-testing task, and the associated increased width facilitates information propagation in the forwarding pass. The newly learnt features combine with existing features in the last layer for meta-learning. Experimental results show that our proposed MAC method outperformed existing ANIL algorithm for non-similar task distribution by approximately 13% (5-shot task setting)
Meta-learning is a branch of machine learning which aims to synthesize data from a distribution of related tasks to efficiently solve new ones. In process control, many systems have similar and well-understood dynamics, which suggests it is feasible to create a generalizable controller through meta-learning. In this work, we formulate a meta reinforcement learning (meta-RL) control strategy that takes advantage of known, offline information for training, such as a model structure. The meta-RL agent is trained over a distribution of model parameters, rather than a single model, enabling the agent to automatically adapt to changes in the process dynamics while maintaining performance. A key design element is the ability to leverage model-based information offline during training, while maintaining a model-free policy structure for interacting with new environments. Our previous work has demonstrated how this approach can be applied to the industrially-relevant problem of tuning proportional-integral controllers to control first order processes. In this work, we briefly reintroduce our methodology and demonstrate how it can be extended to proportional-integral-derivative controllers and second order systems.
In this paper, we develop upon the emerging topic of loss function learning, which aims to learn loss functions that significantly improve the performance of the models trained under them. Specifically, we propose a new meta-learning framework for learning model-agnostic loss functions via a hybrid neuro-symbolic search approach. The framework first uses evolution-based methods to search the space of primitive mathematical operations to find a set of symbolic loss functions. Second, the set of learned loss functions are subsequently parameterized and optimized via an end-to-end gradient-based training procedure. The versatility of the proposed framework is empirically validated on a diverse set of supervised learning tasks. Results show that the meta-learned loss functions discovered by the newly proposed method outperform both the cross-entropy loss and state-of-the-art loss function learning methods on a diverse range of neural network architectures and datasets.
Deep image prior (DIP) is a recently proposed technique for solving imaging inverse problems by fitting the reconstructed images to the output of an untrained convolutional neural network. Unlike pretrained feedforward neural networks, the same DIP can generalize to arbitrary inverse problems, from denoising to phase retrieval, while offering competitive performance at each task. The central disadvantage of DIP is that, while feedforward neural networks can reconstruct an image in a single pass, DIP must gradually update its weights over hundreds to thousands of iterations, at a significant computational cost. In this work we use meta-learning to massively accelerate DIP-based reconstructions. By learning a proper initialization for the DIP weights, we demonstrate a 10x improvement in runtimes across a range of inverse imaging tasks. Moreover, we demonstrate that a network trained to quickly reconstruct faces also generalizes to reconstructing natural image patches.
Early exiting is an effective paradigm for improving the inference efficiency of deep networks. By constructing classifiers with varying resource demands (the exits), such networks allow easy samples to be output at early exits, removing the need for executing deeper layers. While existing works mainly focus on the architectural design of multi-exit networks, the training strategies for such models are largely left unexplored. The current state-of-the-art models treat all samples the same during training. However, the early-exiting behavior during testing has been ignored, leading to a gap between training and testing. In this paper, we propose to bridge this gap by sample weighting. Intuitively, easy samples, which generally exit early in the network during inference, should contribute more to training early classifiers. The training of hard samples (mostly exit from deeper layers), however, should be emphasized by the late classifiers. Our work proposes to adopt a weight prediction network to weight the loss of different training samples at each exit. This weight prediction network and the backbone model are jointly optimized under a meta-learning framework with a novel optimization objective. By bringing the adaptive behavior during inference into the training phase, we show that the proposed weighting mechanism consistently improves the trade-off between classification accuracy and inference efficiency. Code is available at https://github.com/LeapLabTHU/L2W-DEN.
Bilevel optimization (BO) is useful for solving a variety of important machine learning problems including but not limited to hyperparameter optimization, meta-learning, continual learning, and reinforcement learning. Conventional BO methods need to differentiate through the low-level optimization process with implicit differentiation, which requires expensive calculations related to the Hessian matrix. There has been a recent quest for first-order methods for BO, but the methods proposed to date tend to be complicated and impractical for large-scale deep learning applications. In this work, we propose a simple first-order BO algorithm that depends only on first-order gradient information, requires no implicit differentiation, and is practical and efficient for large-scale non-convex functions in deep learning. We provide non-asymptotic convergence analysis of the proposed method to stationary points for non-convex objectives and present empirical results that show its superior practical performance.
Both real and fake news in various domains, such as politics, health, and entertainment are spread via online social media every day, necessitating fake news detection for multiple domains. Among them, fake news in specific domains like politics and health has more serious potential negative impacts on the real world (e.g., the infodemic led by COVID-19 misinformation). Previous studies focus on multi-domain fake news detection, by equally mining and modeling the correlation between domains. However, these multi-domain methods suffer from a seesaw problem: the performance of some domains is often improved at the cost of hurting the performance of other domains, which could lead to an unsatisfying performance in specific domains. To address this issue, we propose a Domain- and Instance-level Transfer Framework for Fake News Detection (DITFEND), which could improve the performance of specific target domains. To transfer coarse-grained domain-level knowledge, we train a general model with data of all domains from the meta-learning perspective. To transfer fine-grained instance-level knowledge and adapt the general model to a target domain, we train a language model on the target domain to evaluate the transferability of each data instance in source domains and re-weigh each instance's contribution. Offline experiments on two datasets demonstrate the effectiveness of DITFEND. Online experiments show that DITFEND brings additional improvements over the base models in a real-world scenario.
As a successful approach to self-supervised learning, contrastive learning aims to learn invariant information shared among distortions of the input sample. While contrastive learning has yielded continuous advancements in sampling strategy and architecture design, it still remains two persistent defects: the interference of task-irrelevant information and sample inefficiency, which are related to the recurring existence of trivial constant solutions. From the perspective of dimensional analysis, we find out that the dimensional redundancy and dimensional confounder are the intrinsic issues behind the phenomena, and provide experimental evidence to support our viewpoint. We further propose a simple yet effective approach MetaMask, short for the dimensional Mask learned by Meta-learning, to learn representations against dimensional redundancy and confounder. MetaMask adopts the redundancy-reduction technique to tackle the dimensional redundancy issue and innovatively introduces a dimensional mask to reduce the gradient effects of specific dimensions containing the confounder, which is trained by employing a meta-learning paradigm with the objective of improving the performance of masked representations on a typical self-supervised task. We provide solid theoretical analyses to prove MetaMask can obtain tighter risk bounds for downstream classification compared to typical contrastive methods. Empirically, our method achieves state-of-the-art performance on various benchmarks.
Packaged fresh-cut lettuce is widely consumed as a major component of vegetable salad owing to its high nutrition, freshness, and convenience. However, enzymatic browning discoloration on lettuce cut edges significantly reduces product quality and shelf life. While there are many research and breeding efforts underway to minimize browning, the progress is hindered by the lack of a rapid and reliable methodology to evaluate browning. Current methods to identify and quantify browning are either too subjective, labor intensive, or inaccurate. In this paper, we report a deep learning model for lettuce browning prediction. To the best of our knowledge, it is the first-of-its-kind on deep learning for lettuce browning prediction using a pretrained Siamese Quadratic Swin (SQ-Swin) transformer with several highlights. First, our model includes quadratic features in the transformer model which is more powerful to incorporate real-world representations than the linear transformer. Second, a multi-scale training strategy is proposed to augment the data and explore more of the inherent self-similarity of the lettuce images. Third, the proposed model uses a siamese architecture which learns the inter-relations among the limited training samples. Fourth, the model is pretrained on the ImageNet and then trained with the reptile meta-learning algorithm to learn higher-order gradients than a regular one. Experiment results on the fresh-cut lettuce datasets show that the proposed SQ-Swin outperforms the traditional methods and other deep learning-based backbones.
Feature extraction methods help in dimensionality reduction and capture relevant information. In time series forecasting (TSF), features can be used as auxiliary information to achieve better accuracy. Traditionally, features used in TSF are handcrafted, which requires domain knowledge and significant data-engineering work. In this research, we first introduce a notion of static and dynamic features, which then enables us to develop our autonomous Feature Retrieving Autoregressive Network for Static features (FRANS) that does not require domain knowledge. The method is based on a CNN classifier that is trained to create for each series a collective and unique class representation either from parts of the series or, if class labels are available, from a set of series of the same class. It allows to discriminate series with similar behaviour but from different classes and makes the features extracted from the classifier to be maximally discriminatory. We explore the interpretability of our features, and evaluate the prediction capabilities of the method within the forecasting meta-learning environment FFORMA. Our results show that our features lead to improvement in accuracy in most situations. Once trained our approach creates features orders of magnitude faster than statistical methods.
This paper presents a method of decoupled pronunciation and prosody modeling to improve the performance of meta-learning-based multilingual speech synthesis. The baseline meta-learning synthesis method adopts a single text encoder with a parameter generator conditioned on language embeddings and a single decoder to predict mel-spectrograms for all languages. In contrast, our proposed method designs a two-stream model structure that contains two encoders and two decoders for pronunciation and prosody modeling, respectively, considering that the pronunciation knowledge and the prosody knowledge should be shared in different ways among languages. In our experiments, our proposed method effectively improved the intelligibility and naturalness of multilingual speech synthesis comparing with the baseline meta-learning synthesis method.
Nowadays, transformer-based models gradually become the default choice for artificial intelligence pioneers. The models also show superiority even in the few-shot scenarios. In this paper, we revisit the classical methods and propose a new few-shot alternative. Specifically, we investigate the few-shot one-class problem, which actually takes a known sample as a reference to detect whether an unknown instance belongs to the same class. This problem can be studied from the perspective of sequence match. It is shown that with meta-learning, the classical sequence match method, i.e. Compare-Aggregate, significantly outperforms transformer ones. The classical approach requires much less training cost. Furthermore, we perform an empirical comparison between two kinds of sequence match approaches under simple fine-tuning and meta-learning. Meta-learning causes the transformer models' features to have high-correlation dimensions. The reason is closely related to the number of layers and heads of transformer models. Experimental codes and data are available at https://github.com/hmt2014/FewOne
The ability to accelerate the design of biological sequences can have a substantial impact on the progress of the medical field. The problem can be framed as a global optimization problem where the objective is an expensive black-box function such that we can query large batches restricted with a limitation of a low number of rounds. Bayesian Optimization is a principled method for tackling this problem. However, the astronomically large state space of biological sequences renders brute-force iterating over all possible sequences infeasible. In this paper, we propose MetaRLBO where we train an autoregressive generative model via Meta-Reinforcement Learning to propose promising sequences for selection via Bayesian Optimization. We pose this problem as that of finding an optimal policy over a distribution of MDPs induced by sampling subsets of the data acquired in the previous rounds. Our in-silico experiments show that meta-learning over such ensembles provides robustness against reward misspecification and achieves competitive results compared to existing strong baselines.
The vision of 5G lies in providing high data rates, low latency (for the aim of near-real-time applications), significantly increased base station capacity, and near-perfect quality of service (QoS) for users, compared to LTE networks. In order to provide such services, 5G systems will support various combinations of access technologies such as LTE, NR, NR-U and Wi-Fi. Each radio access technology (RAT) provides different types of access, and these should be allocated and managed optimally among the users. Besides resource management, 5G systems will also support a dual connectivity service. The orchestration of the network therefore becomes a more difficult problem for system managers with respect to legacy access technologies. In this paper, we propose an algorithm for RAT allocation based on federated meta-learning (FML), which enables RAN intelligent controllers (RICs) to adapt more quickly to dynamically changing environments. We have designed a simulation environment which contains LTE and 5G NR service technologies. In the simulation, our objective is to fulfil UE demands within the deadline of transmission to provide higher QoS values. We compared our proposed algorithm with a single RL agent, the Reptile algorithm and a rule-based heuristic method. Simulation results show that the proposed FML method achieves higher caching rates at first deployment round 21% and 12% respectively. Moreover, proposed approach adapts to new tasks and environments most quickly amongst the compared methods.
Continual learning aims to rapidly and continually learn the current task from a sequence of tasks. Compared to other kinds of methods, the methods based on experience replay have shown great advantages to overcome catastrophic forgetting. One common limitation of this method is the data imbalance between the previous and current tasks, which would further aggravate forgetting. Moreover, how to effectively address the stability-plasticity dilemma in this setting is also an urgent problem to be solved. In this paper, we overcome these challenges by proposing a novel framework called Meta-learning update via Multi-scale Knowledge Distillation and Data Augmentation (MMKDDA). Specifically, we apply multiscale knowledge distillation to grasp the evolution of long-range and short-range spatial relationships at different feature levels to alleviate the problem of data imbalance. Besides, our method mixes the samples from the episodic memory and current task in the online continual training procedure, thus alleviating the side influence due to the change of probability distribution. Moreover, we optimize our model via the meta-learning update resorting to the number of tasks seen previously, which is helpful to keep a better balance between stability and plasticity. Finally, our experimental evaluation on four benchmark datasets shows the effectiveness of the proposed MMKDDA framework against other popular baselines, and ablation studies are also conducted to further analyze the role of each component in our framework.
Causal discovery (CD) from time-varying data is important in neuroscience, medicine, and machine learning. Techniques for CD encompass randomized experiments, which are generally unbiased but expensive, and algorithms such as Granger causality, conditional-independence-based, structural-equation-based, and score-based methods that are only accurate under strong assumptions made by human designers. However, as demonstrated in other areas of machine learning, human expertise is often not entirely accurate and tends to be outperformed in domains with abundant data. In this study, we examine whether we can enhance domain-specific causal discovery for time series using a data-driven approach. Our findings indicate that this procedure significantly outperforms human-designed, domain-agnostic causal discovery methods, such as Mutual Information, VAR-LiNGAM, and Granger Causality on the MOS 6502 microprocessor, the NetSim fMRI dataset, and the Dream3 gene dataset. We argue that, when feasible, the causality field should consider a supervised approach in which domain-specific CD procedures are learned from extensive datasets with known causal relationships, rather than being designed by human specialists. Our findings promise a new approach toward improving CD in neural and medical data and for the broader machine learning community.
Few-shot text classification aims to classify the text under the few-shot scenario. Most of the previous methods adopt optimization-based meta learning to obtain task distribution. However, due to the neglect of matching between the few amount of samples and complicated models, as well as the distinction between useful and useless task features, these methods suffer from the overfitting issue. To address this issue, we propose a novel Adaptive Meta-learner via Gradient Similarity (AMGS) method to improve the model generalization ability to a new task. Specifically, the proposed AMGS alleviates the overfitting based on two aspects: (i) acquiring the potential semantic representation of samples and improving model generalization through the self-supervised auxiliary task in the inner loop, (ii) leveraging the adaptive meta-learner via gradient similarity to add constraints on the gradient obtained by base-learner in the outer loop. Moreover, we make a systematic analysis of the influence of regularization on the entire framework. Experimental results on several benchmarks demonstrate that the proposed AMGS consistently improves few-shot text classification performance compared with the state-of-the-art optimization-based meta-learning approaches.
Recently, due to the poor performance of supervised person re-identification (ReID) to an unseen domain, Domain Generalization (DG) person ReID has attracted a lot of attention which aims to learn a domain-insensitive model and can resist the influence of domain bias. In this paper, we first verify through an experiment that style factors are a vital part of domain bias. Base on this conclusion, we propose a Style Variable and Irrelevant Learning (SVIL) method to eliminate the effect of style factors on the model. Specifically, we design a Style Jitter Module (SJM) in SVIL. The SJM module can enrich the style diversity of the specific source domain and reduce the style differences of various source domains. This leads to the model focusing on identity-relevant information and being insensitive to the style changes. Besides, we organically combine the SJM module with a meta-learning algorithm, maximizing the benefits and further improving the generalization ability of the model. Note that our SJM module is plug-and-play and inference cost-free. Extensive experiments confirm the effectiveness of our SVIL and our method outperforms the state-of-the-art methods on DG-ReID benchmarks by a large margin.
graph neural networks (GNNs) are the dominant paradigm for modeling and handling graph structure data by learning universal node representation. The traditional way of training GNNs depends on a great many labeled data, which results in high requirements on cost and time. In some special scene, it is even unavailable and impracticable. Self-supervised representation learning, which can generate labels by graph structure data itself, is a potential approach to tackle this problem. And turning to research on self-supervised learning problem for heterogeneous graphs is more challenging than dealing with homogeneous graphs, also there are fewer studies about it. In this paper, we propose a SElfsupervised learning method for heterogeneous graph via Structure Information based on Metapath (SESIM). The proposed model can construct pretext tasks by predicting jump number between nodes in each metapath to improve the representation ability of primary task. In order to predict jump number, SESIM uses data itself to generate labels, avoiding time-consuming manual labeling. Moreover, predicting jump number in each metapath can effectively utilize graph structure information, which is the essential property between nodes. Therefore, SESIM deepens the understanding of models for graph structure. At last, we train primary task and pretext tasks jointly, and use meta-learning to balance the contribution of pretext tasks for primary task. Empirical results validate the performance of SESIM method and demonstrate that this method can improve the representation ability of traditional neural networks on link prediction task and node classification task.