Meta-learning Scheme Research Articles

Out-of-Domain Generalization From a Single Source: An Uncertainty Quantification Approach.

We are concerned with a worst-case scenario in model generalization, in the sense that a model aims to perform well on many unseen domains while there is only one single domain available for training. We propose Meta-Learning based Adversarial Domain Augmentation to solve this Out-of-Domain generalization problem. The key idea is to leverage adversarial training to create "fictitious" yet "challenging" populations, from which a model can learn to generalize with theoretical guarantees. To facilitate fast and desirable domain augmentation, we cast the model training in a meta-learning scheme and use a Wasserstein Auto-Encoder to relax the widely used worst-case constraint. We further improve our method by integrating uncertainty quantification for efficient domain generalization. Extensive experiments on multiple benchmark datasets indicate its superior performance in tackling single domain generalization.

Few-Shot Calibration of Set Predictors via Meta-Learned Cross-Validation-Based Conformal Prediction.

Conventional frequentist learning is known to yield poorly calibrated models that fail to reliably quantify the uncertainty of their decisions. Bayesian learning can improve calibration, but formal guarantees apply only under restrictive assumptions about correct model specification. Conformal prediction (CP) offers a general framework for the design of set predictors with calibration guarantees that hold regardless of the underlying data generation mechanism. However, when training data are limited, CP tends to produce large, and hence uninformative, predicted sets. This paper introduces a novel meta-learning solution that aims at reducing the set prediction size. Unlike prior work, the proposed meta-learning scheme, referred to as meta-XB, i) builds on cross-validation-based CP, rather than the less efficient validation-based CP; and ii) preserves formal per-task calibration guarantees, rather than less stringent task-marginal guarantees. Finally, meta-XB is extended to adaptive non-conformal scores, which are shown empirically to further enhance marginal per-input calibration.

Joint Optimization of Class-Specific Training- and Test-Time Data Augmentation in Segmentation.

This paper presents an effective and general data augmentation framework for medical image segmentation. We adopt a computationally efficient and data-efficient gradient-based meta-learning scheme to explicitly align the distribution of training and validation data which is used as a proxy for unseen test data. We improve the current data augmentation strategies with two core designs. First, we learn class-specific training-time data augmentation (TRA) effectively increasing the heterogeneity within the training subsets and tackling the class imbalance common in segmentation. Second, we jointly optimize TRA and test-time data augmentation (TEA), which are closely connected as both aim to align the training and test data distribution but were so far considered separately in previous works. We demonstrate the effectiveness of our method on four medical image segmentation tasks across different scenarios with two state-of-the-art segmentation models, DeepMedic and nnU-Net. Extensive experimentation shows that the proposed data augmentation framework can significantly and consistently improve the segmentation performance when compared to existing solutions. Code is publicly available1.

Online Meta-Learning for Hybrid Model-Based Deep Receivers

Recent years have witnessed growing interest in the application of deep neural networks (DNNs) for receiver design, which can potentially be applied in complex environments without relying on knowledge of the channel model. However, the dynamic nature of communication channels often leads to rapid distribution shifts, which may require periodically retraining. This paper formulates a data-efficient two-stage training method that facilitates rapid online adaptation. Our training mechanism uses a predictive meta-learning scheme to train rapidly from data corresponding to both current and past channel realizations. Our method is applicable to any deep neural network (DNN)-based receiver, and does not require transmission of new pilot data for training. To illustrate the proposed approach, we study DNN-aided receivers that utilize an interpretable model-based architecture, and introduce a modular training strategy based on predictive meta-learning. We demonstrate our techniques in simulations on a synthetic linear channel, a synthetic non-linear channel, and a COST 2100 channel. Our results demonstrate that the proposed online training scheme allows receivers to outperform previous techniques based on self-supervision and joint-learning by a margin of up to 2.5 dB in coded bit error rate in rapidly-varying scenarios.

Meta-Machine-Learning-Based Quantum Scar Detector

A remarkable phenomenon in contemporary physics is quantum scarring in systems whose classical dynamics are chaotic, where certain wave functions tend to concentrate on classical periodic orbits of low periods. Quantum scarring has been studied for more than four decades, but detecting quantum scars still mostly relies on human visualization of the wave-function patterns. The widespread and successful applications of machine learning in many branches of physics suggest the possibility of using artificial neural networks for automated detection of quantum scars. Conventional machine learning often requires substantial training data, but, for quantum scars, this poses a significant challenge: in typical systems the available distinct quantum scarring states are rare. We develop a meta machine-learning approach to accurately detect quantum scars in a fully automated and highly efficient fashion. In particular, taking advantage of some standard large datasets such as Omniglot from the field of image classification, we train a ``preliminary'' version of the neural network that has the ability to distinguish different classes of noisy images. We then perform few-shot classification to further train the neural network but with a small number of quantum scars. We demonstrate that the meta-learning scheme can find the correct quantum scars from thousands of images of wave functions without any human intervention, regardless of the symmetry of the underlying system. From a general applied point of view, our success opens the door to exploiting meta learning for solving challenging image detection and classification problems in other fields of science and engineering. For example, in microlasing systems, identifying scarring states is critical as these states are desired for directional emission. The task is also important for quantum-dot devices where the scarring states can lead to resonances in the conductance.

Magi-Net: Meta Negative Network for Early Activity Prediction.

Early activity prediction/recognition aims to recognize action categories before they are fully conveyed. Compared to full-length action sequences, partial video sequences only provide insufficient discrimination information, which makes predicting the class labels for some similar activities challenging, especially when only very few frames can be observed. To address this challenge, in this paper, we propose a novel meta negative network, namely, Magi-Net, that utilizes a contrastive learning scheme to alleviate the insufficiency of discriminative information. In our Magi-Net model, the positive samples are generated by augmenting an input anchor conditioned on all observation ratios, while the negative samples are selected from a trainable negative look-up memory (LUM) table, which stores the training samples and the corresponding misleading categories. Furthermore, a meta negative sample optimization strategy (MetaSOS) is proposed to boost the training of Magi-Net by encouraging the model to learn from the most informative negative samples via a meta learning scheme. Extensive experiments are conducted on several public skeleton-based activity datasets, and the results show the efficacy of the proposed Magi-Net model.

Auto-Encoder Based Orthogonal Time Frequency Space Modulation and Detection With Meta-Learning

To tackle a Doppler sensitivity problem of orthogonal frequency division multiplexing (OFDM), orthogonal time frequency space (OTFS) has been investigated, where information is carried over delay-Doppler domain. In this paper, to improve communication reliability in doubly dispersive channel, an auto-encoder (AE)-based OTFS modulation and detection scheme is developed, where the transmit OTFS waveform and its associated detection scheme at the receiver are jointly optimized in a deep learning framework. However, the conventional AE architecture which takes one-hot encoded input vector is hard to be reused in OTFS due to its enormous input dimensionality that increases exponentially on the number of grid points in delay-Doppler domain. To overcome it, we divide the delay-Doppler grid into multiple subblocks and associate the one-hot encoded vector with each subblock. Then, by concatenating them, one multi-hot vector is formed and exploited as the input vector for the proposed AE-based OTFS modulation and detection. We also develop a meta-learning scheme to effectively train the AE-based OTFS transceiver for newly updated channel profile.

REIN-2: Giving birth to prepared reinforcement learning agents using reinforcement learning agents

Deep Reinforcement Learning (Deep RL) has been in the spotlight for the past few years, due to its remarkable abilities to solve problems which were considered to be practically unsolvable using traditional Machine Learning methods. However, even state-of-the-art Deep RL algorithms have various weaknesses that prevent them from being used extensively within industry applications, with one such major weakness being their sample-inefficiency. In an effort to patch these issues, we integrated a meta-learning technique in order to shift the objective of learning to solve a task into the objective of learning how to learn to solve a task (or a set of tasks), which we empirically show that improves overall stability and performance of Deep RL algorithms. Our model, named REIN-2, is a meta-learning scheme formulated within the RL framework, the goal of which is to develop a meta-RL agent (meta-learner) that learns how to produce other RL agents (inner-learners) that are capable of solving given environments. For this task, we convert the typical interaction of an RL agent with the environment into a new, single environment for the meta-learner to interact with. Compared to traditional state-of-the-art Deep RL algorithms, experimental results show remarkable performance of our model in popular OpenAI Gym environments in terms of scoring and sample efficiency, including the Mountain Car hard-exploration environment.

NOSnoop: An Effective Collaborative Meta-Learning Scheme Against Property Inference Attack

Collaborative learning has been used to train a joint model on geographically diverse data through periodically sharing knowledge. Although participants keep the data locally in collaborative learning, the adversary can still launch inference attacks through participants’ shared information. In this paper, we focus on the property inference attack during model training and design a novel defense mechanism, namely NOSnoop, to defend such an attack. We propose a collaborative meta-learning architecture to learn the common knowledge over all participants and utilize the natural advantage of meta-learning to hide the sensitive property data. We consider both irrelevant property and relevant property preservation in NOSnoop. For irrelevant property preservation, we utilize the inherent advantage of meta-learning to hide the sensitive property data in meta-training support dataset. Thus, the adversary cannot capture the key information related to the sensitive properties and cannot infer victim’s private property successfully. For relevant property preservation, an adversarial game is further proposed to reduce the inference success rate of the adversary. We conduct comprehensive experiments to evaluate the effectiveness of NOSnoop. When hiding the sensitive property data in meta-training support dataset, NOSnoop achieves an inference AUC score as low as 0.4984 for irrelevant property preservation, meaning the adversary cannot distinguish whether the training batch has the sensitive property data or not. When preserving the relevant property, NOSnoop is able to achieve an inference AUC score of 0.5091 without compromising model utility.

A dual meta-learning framework based on idle data for enhancing segmentation of pancreatic cancer.

Automated segmentation of pancreatic cancer is vital for clinical diagnosis and treatment. However, the small size and inconspicuous boundaries limit the segmentation performance, which is further exacerbated for deep learning techniques with the few training samples due to the high threshold of image acquisition and annotation. To alleviate this issue caused by the small-scale dataset, we collect idle multi-parametric MRIs of pancreatic cancer from different studies to construct a relatively large dataset for enhancing the CT pancreatic cancer segmentation. Therefore, we propose a deep learning segmentation model with the dual meta-learning framework for pancreatic cancer. It can integrate the common knowledge of tumors obtained from idle MRIs and salient knowledge from CT images, making high-level features more discriminative. Specifically, the random intermediate modalities between MRIs and CT are first generated to smoothly fill in the gaps in visual appearance and provide rich intermediate representations for ensuing meta-learning scheme. Subsequently, we employ intermediate modalities-based model-agnostic meta-learning to capture and transfer commonalities. At last, a meta-optimizer is utilized to adaptively learn the salient features within CT data, thus alleviating the interference due to internal differences. Comprehensive experimental results demonstrated our method achieved the promising segmentation performance, with a max Dice score of 64.94% on our private dataset, and outperformed state-of-the-art methods on a public pancreatic cancer CT dataset. The proposed method is an effective pancreatic cancer segmentation framework, which can be easily integrated into other segmentation networks and thus promises to be a potential paradigm for alleviating data scarcity challenges using idle data.

Federated Meta-Learning Enhanced Acoustic Radio Cooperative Framework for Ocean of Things

Ocean of Things, consisting of multiple buoys distributed on the sea, is an acoustic radio cooperative wireless network that aims to acquire underwater information. In this paper, a deep neural network (DNN)-based receiver with data augmentation, termed chirp (C)-DNN, is developed for a buoy that uses chirp modulation-based underwater acoustic communications. To further solve the problem that the training data at a single buoy may not be sufficient, a federated meta-learning (FML) scheme is proposed to train the DNN-based receiver by exploiting the model parameters from multiple buoys. We analyze the convergence performance of FML and derive a closed-form expression for the convergence rate, accounting for the impacts of scheduling ratios, local epochs, and data volumes on a single node. Simulation results show that the proposed C-DNN receiver that is trained with sufficient data achieves better bit error rate performance and lower complexity than classical matched filter (MF)-based detectors. The proposed FML also outperforms the MF-based method after several communication rounds and achieves better generalization performance than federated learning based systems.

Robust Spike-Based Continual Meta-Learning Improved by Restricted Minimum Error Entropy Criterion.

The spiking neural network (SNN) is regarded as a promising candidate to deal with the great challenges presented by current machine learning techniques, including the high energy consumption induced by deep neural networks. However, there is still a great gap between SNNs and the online meta-learning performance of artificial neural networks. Importantly, existing spike-based online meta-learning models do not target the robust learning based on spatio-temporal dynamics and superior machine learning theory. In this invited article, we propose a novel spike-based framework with minimum error entropy, called MeMEE, using the entropy theory to establish the gradient-based online meta-learning scheme in a recurrent SNN architecture. We examine the performance based on various types of tasks, including autonomous navigation and the working memory test. The experimental results show that the proposed MeMEE model can effectively improve the accuracy and the robustness of the spike-based meta-learning performance. More importantly, the proposed MeMEE model emphasizes the application of the modern information theoretic learning approach on the state-of-the-art spike-based learning algorithms. Therefore, in this invited paper, we provide new perspectives for further integration of advanced information theory in machine learning to improve the learning performance of SNNs, which could be of great merit to applied developments with spike-based neuromorphic systems.

Resilient UAV Swarm Communications With Graph Convolutional Neural Network

In this paper, we study the self-healing problem of unmanned aerial vehicle (UAV) swarm network (USNET) that is required to quickly rebuild the communication connectivity under unpredictable external destructions (UEDs). Firstly, to cope with the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">one-off UEDs</i> , we propose a graph convolutional neural network (GCN) that can find the recovery topology of the USNET in an on-line manner. Secondly, to cope with <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">general UEDs</i> , we develop a GCN based trajectory planning algorithm that can make UAVs rebuild the communication connectivity during the self-healing process. We also design a meta learning scheme to facilitate the on-line executions of the GCN. Numerical results show that the proposed algorithms can rebuild the communication connectivity of the USNET more quickly than the existing algorithms under both one-off UEDs and general UEDs. The simulation results also show that the meta learning scheme can not only enhance the performance of the GCN but also reduce the time complexity of the on-line executions.

Meta PID Attention Network for Flexible and Efficient Real-World Noisy Image Denoising.

Recent deep convolutional neural networks for real-world noisy image denoising have shown a huge boost in performance by training a well-engineered network over external image pairs. However, most of these methods are generally trained with supervision. Once the testing data is no longer compatible with the training conditions, they can exhibit poor generalization and easily result in severe overfitting or degrading performances. To tackle this barrier, we propose a novel denoising algorithm, dubbed as Meta PID Attention Network (MPA-Net). Our MPA-Net is built based upon stacking Meta PID Attention Modules (MPAMs). In each MPAM, we utilize a second-order attention module (SAM) to exploit the channel-wise feature correlations with second-order statistics, which are then adaptively updated via a proportional-integral-derivative (PID) guided meta-learning framework. This learning framework exerts the unique property of the PID controller and meta-learning scheme to dynamically generate filter weights for beneficial update of the extracted features within a feedback control system. Moreover, the dynamic nature of the framework enables the generated weights to be flexibly tweaked according to the input at test time. Thus, MPAM not only achieves discriminative feature learning, but also facilitates a robust generalization ability on distinct noises for real images. Extensive experiments on ten datasets are conducted to inspect the effectiveness of the proposed MPA-Net quantitatively and qualitatively, which demonstrates both its superior denoising performance and promising generalization ability that goes beyond those of the state-of-the-art denoising methods.

ROA: A Rapid Learning Scheme for In-Situ Memristor Networks.

Memristors show great promise in neuromorphic computing owing to their high-density integration, fast computing and low-energy consumption. However, the non-ideal update of synaptic weight in memristor devices, including nonlinearity, asymmetry and device variation, still poses challenges to the in-situ learning of memristors, thereby limiting their broad applications. Although the existing offline learning schemes can avoid this problem by transferring the weight optimization process into cloud, it is difficult to adapt to unseen tasks and uncertain environments. Here, we propose a bi-level meta-learning scheme that can alleviate the non-ideal update problem, and achieve fast adaptation and high accuracy, named Rapid One-step Adaption (ROA). By introducing a special regularization constraint and a dynamic learning rate strategy for in-situ learning, the ROA method effectively combines offline pre-training and online rapid one-step adaption. Furthermore, we implemented it on memristor-based neural networks to solve few-shot learning tasks, proving its superiority over the pure offline and online schemes under noisy conditions. This method can solve in-situ learning in non-ideal memristor networks, providing potential applications of on-chip neuromorphic learning and edge computing.

Learning to Attack Real-World Models for Person Re-identification via Virtual-Guided Meta-Learning

Recent advances in person re-identification (re-ID) have led to impressive retrieval accuracy. However, existing re-ID models are challenged by the adversarial examples crafted by adding quasi-imperceptible perturbations. Moreover, re-ID systems face the domain shift issue that training and testing domains are not consistent. In this study, we argue that learning powerful attackers with high universality that works well on unseen domains is an important step in promoting the robustness of re-ID systems. Therefore, we introduce a novel universal attack algorithm called ``MetaAttack'' for person re-ID. MetaAttack can mislead re-ID models on unseen domains by a universal adversarial perturbation. Specifically, to capture common patterns across different domains, we propose a meta-learning scheme to seek the universal perturbation via the gradient interaction between meta-train and meta-test formed by two datasets. We also take advantage of a virtual dataset (PersonX), instead of real ones, to conduct meta-test. This scheme not only enables us to learn with more comprehensive variation factors but also mitigates the negative effects caused by biased factors of real datasets. Experiments on three large-scale re-ID datasets demonstrate the effectiveness of our method in attacking re-ID models on unseen domains. Our final visualization results reveal some new properties of existing re-ID systems, which can guide us in designing a more robust re-ID model. Code and supplemental material are available at \url{https://github.com/FlyingRoastDuck/MetaAttack_AAAI21}.

A Meta-Learning Scheme for Adaptive Short-Term Network Traffic Prediction

Network traffic prediction is a fundamental prerequisite for dynamic resource provisioning in wireline and wireless networks, but is known to be challenging due to non-stationarity and due to its burstiness and self-similar nature. The prediction of network traffic at the user level is particularly challenging, because the traffic characteristics emerge from a complex interaction of user level and application protocol behavior. In this work we address the problem of predicting the network traffic at the user level over a short horizon, motivated by its applications in cellular scheduling. Motivated by recent works on robust adversarial learning, we treat the prediction problem for non-stationary traffic in an adversarial context, and propose a meta-learning scheme that consists of a set of predictors, each optimized to predict a particular kind of traffic, and of a master policy that is trained for choosing the best fit predictor dynamically based on recent prediction performance, using deep reinforcement learning. We evaluate the proposed meta-learning scheme on a variety of traffic traces consisting of video and non-video traffic. Our results show that it consistently outperforms state-of-the-art predictors, and can adapt to before unseen traffic without the need for retraining the individual predictors.

ConvProtoNet: Deep Prototype Induction towards Better Class Representation for Few-Shot Malware Classification

Traditional malware classification relies on known malware types and significantly large datasets labeled manually which limits its ability to recognize new malware classes. For unknown malware types or new variants of existing malware containing only a few samples each class, common classification methods often fail to work well due to severe overfitting. In this paper, we propose a new neural network structure called ConvProtoNet which employs few-shot learning to address the problem of scarce malware samples while prevent from overfitting. We design a convolutional induction module to replace the insufficient prototype reduction in most few-shot models and generates more appropriate class-level malware prototypes for classification. We also adopt meta-learning scheme to make classifier robust enough to adapt unseen malware classes without fine-tuning. Even in extreme conditions where only 5 samples in each class are provided, ConvProtoNet still achieves more than 70% accuracy on average and outperforms other traditional malware classification methods or existed few-shot models in experiments conducted on several datasets. Extra experiments across datasets illustrate that ConvProtoNet learns general knowledge of malware which is dataset-invariant and careful model analysis proves effectiveness of ConvProtoNet in few-shot malware classification.

Ensemble Learning from Crowds

Traditional learning from crowdsourced labeled data consists of two stages: inferring true labels for instances from their multiple noisy labels and building a learning model using these instances with the inferred labels. This straightforward two-stage learning scheme suffers from two weaknesses: (1) the accuracy of inference may be very low; (2) useful information may be lost during inference. In this paper, we proposed a novel ensemble method for learning from crowds. Our proposed method is a meta-learning scheme. It first uses a bootstrapping process to create $M$M sub-datasets from an original crowdsourced labeled dataset. For each sub-dataset, each instance is duplicated with different weights according to the distribution and class memberships of its multiple noisy labels. A base classifier is then trained from this extended sub-dataset. Finally, unlabeled instances are predicted by aggregating the outputs of these $M$M base classifiers. Because the proposed method gets rid of the inference procedure and uses the full dataset to train learning models, it preserves the useful information for learning as much as possible. Experimental results on nine simulated and two real-world crowdsourcing datasets consistently show that the proposed ensemble learning method significantly outperforms five state-of-the-art methods.

Cost-sensitive meta-learning classifiers: MEPAR-miner and DIFACONN-miner

Cost-sensitive learning which deals with classification problems that have non-uniform costs has attracted great attention from the machine learning and data mining communities in recent years. In this study, a rescaling based meta-learning scheme is applied to cost-insensitive MEPAR-miner and DIFACONN-miner algorithms which were previously developed by the authors in order to make the algorithms cost-sensitive. Rescaling is realized in two ways by means of oversampling and undersampling by resampling the training instances in proportion to their costs. The proposed algorithms can extract rules for both binary and n-ary classification problems and also handle data sets that have missing values. An extensive computational study is performed on different types of classification benchmark problems with the aim of testing the performances of the algorithms. Comparisons with traditional cost-sensitive meta-learning algorithms and cost-insensitive MEPAR-miner and DIFACONN-miner algorithms show that the proposed cost-sensitive algorithms are competitive meta-learning algorithms and able to produce accurate and effective classification rules with low misclassification costs.