Search

Meta-learning empowers learning systems with the ability to acquire knowledge from multiple tasks, enabling faster adaptation and generalization to new tasks. This review provides a comprehensive technical overview of meta-learning, emphasizing its importance in real-world applications where data may be scarce or expensive to obtain. The article covers the state-of-the-art meta-learning approaches and explores the relationship between meta-learning and multi-task learning, transfer learning, domain adaptation and generalization, self-supervised learning, personalized federated learning, and continual learning. By highlighting the synergies between these topics and the field of meta-learning, the article demonstrates how advancements in one area can benefit the field as a whole, while avoiding unnecessary duplication of efforts. Additionally, the article delves into advanced meta-learning topics such as learning from complex multi-modal task distributions, unsupervised meta-learning, learning to efficiently adapt to data distribution shifts, and continual meta-learning. Lastly, the article highlights open problems and challenges for future research in the field. By synthesizing the latest research developments, this article provides a thorough understanding of meta-learning and its potential impact on various machine learning applications. We believe that this technical overview will contribute to the advancement of meta-learning and its practical implications in addressing real-world problems.

Federated learning has emerged as a promising approach for training machine learning models on decentralized data sources while preserving data privacy. However, challenges such as communication bottlenecks, heterogeneity of client devices, and non-i.i.d. data distribution pose significant obstacles to achieving optimal model performance. We propose a novel framework that combines federated learning with meta-learning techniques to enhance both efficiency and generalization capabilities. Our approach introduces a federated modulator that learns contextual information from data batches and uses this knowledge to generate modulation parameters. These parameters dynamically adjust the activations of a base model, which operates using a MAML-based approach for model personalization. Experimental results across diverse datasets highlight the improvements in convergence speed and model performance compared to existing federated learning approaches. These findings highlight the potential of incorporating contextual information and meta-learning techniques into federated learning, paving the way for advancements in distributed machine learning paradigms. Copyright © 2024 by SIAM.

Meta-learning or learning to learn involves training a model on various learning tasks in a way that allows it to quickly learn new tasks from the same distribution using only a small amount of training data (i.e., few-shot learning). Current meta-learning methods implicitly assume that the distribution over tasks is unimodal and consists of tasks belonging to a common domain, which significantly reduces the variety of task distributions they can handle. However, in real-world applications, tasks are often very diverse and come from multiple different domains, making it challenging to meta-learn common knowledge shared across the entire task distribution. In this paper, we propose a method for meta-learning from a multimodal task distribution. The proposed method learns multiple sets of meta-parameters (acting as different initializations of a neural network model) and uses a task encoder to select the best initialization to fine-tune for a new task. More specifically, with a few training examples from a task sampled from an unknown mode, the proposed method predicts which set of meta-parameters (i.e., model’s initialization) would lead to a fast adaptation and a good post-adaptation performance on that task. We evaluate the proposed method on a diverse set of few-shot regression and image classification tasks. The results demonstrate the superiority of the proposed method compared to other state of-the-art meta-learning methods and the benefit of learning multiple model initializations when tasks are sampled from a multimodal task distribution. © 2023 IEEE.

In most industries, the working conditions of equipment vary significantly from one site to another, from one time of a year to another, and so on. This variation poses a severe challenge for data-driven fault identification methods: it introduces a change in the data distribution. This contradicts the underlying assumption of most machine learning methods, namely that training and test samples follow the same distribution. Domain Adaptation (DA) methods aim to address this problem by minimizing the distribution distance between training (source) and test (target) samples. However, in the area of predictive maintenance, this idea is complicated by the fact that different classes – fault categories – also vary across domains. Most of the state-of-the-art DA methods assume that the data in the target domain is complete, i.e., that we have access to examples from all the possible classes or faulty categories during adaptation. In reality, this is often very difficult to guarantee. Therefore, there is a need for a domain adaptation method that is able to align the source and target domains even in cases of having access to an incomplete set of test data. This paper presents our work in progress as we propose an approach for such a setting based on maintaining the geometry information of source samples during the adaptation. This way, the model can capture the relationships between different fault categories and preserve them in the constructed domain-invariant feature space, even in situations where some classes are entirely missing. This paper examines this idea using artificial data sets to demonstrate the effectiveness of geometry-preserving transformation. We have also started investigations on real-world predictive maintenance datasets, such as CWRU.

Domain Adaptation (DA) aims to transfer knowledge from a source to a target domain by aligning their respective data distributions. In the unsupervised setting, however, this may cause the source and target samples of different classes to align to each other, consequently leading to negative transfer. Semi-Supervised Domain Adaptation (SSDA) tries to solve such class misalignment problem by exploiting a few sample labels in the target domain. This paper proposes a new SSDA method called Adversarial Contrastive Semi-Supervised Domain Adaptation (ACSSDA) which combines two objectives, optimized for the case where very few target sample labels are available, to learn a shared feature representation for both source and target domains. ACSSDA uses a domain classifier to ensure that the resulting feature space is domain agnostic. Simultaneously, Contrastive loss aims to pull together samples of the same class and push apart samples of different classes. This is shown to reduce class misalignment and negative transfer even with as little as a single labeled sample per class. We demonstrate the effectiveness of ACSSDA with experiments on several benchmark data sets. The results show the superiority of our method over state-of-the-art approaches., As manuscript in Thesis.

In contextual anomaly detection, an object is only considered anomalous within a specific context. Most existing methods use a single context based on a set of user-specified contextual features. However, identifying the right context can be very challenging in practice, especially in datasets with a large number of attributes. Furthermore, in real-world systems, there might be multiple anomalies that occur in different contexts and, therefore, require a combination of several "useful" contexts to unveil them. In this work, we propose a novel approach, called WisCon (Wisdom of the Contexts), to effectively detect complex contextual anomalies in situations where the true contextual and behavioral attributes are unknown. Our method constructs an ensemble of multiple contexts, with varying importance scores, based on the assumption that not all useful contexts are equally so. We estimate the importance of each context using an active learning approach with a novel query strategy. Experiments show that WisCon significantly outperforms existing baselines in different categories (i.e., active classifiers, unsupervised contextual, and non-contextual anomaly detectors) on 18 datasets. Furthermore, the results support our initial hypothesis that there is no single perfect context that successfully uncovers all kinds of contextual anomalies, and leveraging the "wisdom" of multiple contexts is necessary. © 2022, The Author(s)., Som manuskript i avhandling / As manuscript in thesis

This work proposes a new exchangeability test for a random sequence through a martingale-based approach. Its main contributions include 1) an additive martingale which is more amenable for designing exchangeability tests by exploiting the Hoeffding-Azuma lemma and 2) different betting functions for constructing the additive martingale. By choosing the underlying probability density function of p-values as a betting function, it can be shown that, when a change-point appears, a satisfying trade-off between the smoothness and expected one-step increment of the martingale sequence can be obtained. An online algorithm based on beta distribution parametrization for constructing this betting function is discussed in detail as well. © 2021, IGI Global.

Extracting operation cycles from the historical reading of sensors is an essential step in IoT data analytics. For instance, we can exploit the obtained cycles for learning the normal states to feed into semi-supervised models or dictionaries for efficient real-time anomaly detection on the sensors. However, this is a difficult problem due to this fact that we may have different types of cycles, each of which with varying lengths. Current approaches are highly dependent on manual efforts by the aid of visualization and knowledge of domain experts, which is not feasible on a large scale. We propose a fully automated method called CycleFootprint that can: 1) identify the most relevant signal that has the most obvious recurring patterns among multiple signals; and 2) automatically find the cycles from the selected signal. The main idea behind CycleFootprint is mining footprints in the cycles. We assume that there should be a unique pattern in each cycle that shows up repeatedly in each cycle. By mining those footprints, we can identify cycles. We evaluate our method with existing labeled ground truth data of a real separator in marine application equipped with multiple health monitoring sensors. 86\% of cycles extracted by our method match fully or with at least 99\% overlap with true cycles, which sounds promising given its unsupervised and fully automated nature. © Springer Nature Switzerland AG 2020

Human activity recognition has become an active research field over the past few years due to its wide application in various fields such as health-care, smart home monitoring, and surveillance. Existing approaches for activity recognition in smart homes have achieved promising results. Most of these approaches evaluate real-time recognition of activities using only sensor activations that precede the evaluation time (where the decision is made). However, in several critical situations, such as diagnosing people with dementia, “preceding sensor activations” are not always sufficient to accurately recognize the inhabitant's daily activities in each evaluated time. To improve performance, we propose a method that delays the recognition process in order to include some sensor activations that occur after the point in time where the decision needs to be made. For this, the proposed method uses multiple incremental fuzzy temporal windows to extract features from both preceding and some oncoming sensor activations. The proposed method is evaluated with two temporal deep learning models (convolutional neural network and long short-term memory), on a binary sensor dataset of real daily living activities. The experimental evaluation shows that the proposed method achieves significantly better results than the real-time approach, and that the representation with fuzzy temporal windows enhances performance within deep learning models. © Copyright 2020 IEEE, Other funding: Marie Sklodowska-Curie EU Framework for Research

This paper demonstrates how to explore and visualize different types of structure in data, including clusters, anomalies, causal relations, and higher order relations. The methods are developed with the goal of being as automatic as possible and applicable to massive, streaming, and distributed data. Finally, a decentralized learning scheme is discussed, enabling finding structure in the data without collecting the data centrally., BIDAF

We approach the problem of identifying and interpreting clusters over different time scales and granularity in multivariate time series data. We extract statistical features over a sliding window of each time series, and then use a Gaussian mixture model to identify clusters which are then projected back on the data streams. The human analyst can then further analyze this projection and adjust the size of the sliding window and the number of clusters in order to capture the different types of clusters over different time scales. We demonstrate the effectiveness of our approach in two different application scenarios: (1) fleet management and (2) district heating, wherein each scenario, several different types of meaningful clusters can be identified when varying over these dimensions. © 2019 Copyright held by the owner/author(s). Publication rights licensed to ACM.