Distributed Learning Research Articles

Ambiguity-aware breast tumor cellularity estimation via self-ensemble label distribution learning.

In this work, we address the task of tumor cellularity (TC) estimation with a novel framework based on the label distribution learning (LDL) paradigm. We propose a self-ensemble label distribution learning framework (SLDL) to resolve the challenges of existing LDL-based methods, including difficulties for inter-rater ambiguity exploitation, proper and flexible label distribution generation, and accurate TC value recovery. The proposed SLDL makes four main contributions which have been demonstrated to be quite effective in numerous experiments. First, we propose an expertness-aware conditional VAE for diversified single-rater modeling and an attention-based multi-rater fusion strategy that enables effective inter-rater ambiguity exploitation. Second, we propose a template-based label distribution generation method that is tailored for the TC estimation task and constructs label distributions based on the annotation priors. Third, we propose a novel restricted distribution loss, significantly improving the TC value estimation by effectively regularizing the learning with unimodal loss and regression loss. Fourth, to the best of our knowledge, we are the first to simultaneously leverage inter-rater and intra-rater variability to address the label ambiguity issue in the breast tumor cellularity estimation tasks. The experimental results on the public BreastPathQ dataset demonstrate that the SLDL outperforms the existing methods by a large margin and achieves new state-of-the-art results in the TC estimation task. The code will be available from https://github.com/PerceptionComputingLab/ULTRA.

DR-GAN: Distribution Regularization for Text-to-Image Generation.

This article presents a new text-to-image (T2I) generation model, named distribution regularization generative adversarial network (DR-GAN), to generate images from text descriptions from improved distribution learning. In DR-GAN, we introduce two novel modules: a semantic disentangling module (SDM) and a distribution normalization module (DNM). SDM combines the spatial self-attention mechanism (SSAM) and a new semantic disentangling loss (SDL) to help the generator distill key semantic information for the image generation. DNM uses a variational auto-encoder (VAE) to normalize and denoise the image latent distribution, which can help the discriminator better distinguish synthesized images from real images. DNM also adopts a distribution adversarial loss (DAL) to guide the generator to align with normalized real image distributions in the latent space. Extensive experiments on two public datasets demonstrated that our DR-GAN achieved a competitive performance in the T2I task. The code link: https://github.com/Tan-H-C/DR-GAN-Distribution-Regularization-for-Text-to-Image-Generation.

Nash Equilibrium Seeking Algorithm Design for Distributed Nonsmooth Multicluster Games Over Weight-Balanced Digraphs.

In this article, we study the multicluster games over weight-balanced digraphs, where the cost functions of all players are nonsmooth. Besides, in the problem, not only are the decisions of all players constrained by heterogeneous local constraints but also the decisions of players in the same cluster are constrained by coupling constraints. Due to the nonsmooth cost functions, the coupling constraints, the general local convex constraints, and the weight-balanced digraphs, existing Nash equilibrium seeking algorithms cannot solve our problem. In order to seek the Nash equilibrium of the game, we design a distributed algorithm based on subgradient descent, differential inclusions, and projection operations. In the algorithm, a distributed learning strategy is embedded for the players to estimate the decisions of other players. Moreover, we analyze the asymptotical convergence of the algorithm via set-valued LaSalle invariance principle. Finally, a numerical simulation about electricity market games is presented to illustrate the effectiveness of our result.

Predicting Label Distribution From Tie-Allowed Multi-Label Ranking.

Label distribution offers more information about label polysemy than logical label. There are presently two approaches to obtaining label distributions: LDL (label distribution learning) and LE (label enhancement). In LDL, experts must annotate training instances with label distributions, and a predictive function is trained on this training set to obtain label distributions. In LE, experts must annotate instances with logical labels, and label distributions are recovered from them. However, LDL is limited by expensive annotations, and LE has no performance guarantee. Therefore, we investigate how to predict label distribution from TMLR (tie-allowed multi-label ranking) which is a compromise on annotation cost but has good performance guarantees. On the one hand, we theoretically dissect the relationship between TMLR and label distribution. We define EAE (expected approximation error) to quantify the quality of an annotation, provide EAE bounds for TMLR, and derive the optimal range of label distributions corresponding to a given TMLR annotation. On the other hand, we propose a framework for predicting label distribution from TMLR via conditional Dirichlet mixtures. This framework blends the procedures of recovering and learning label distributions end-to-end and allows us to effortlessly encode our knowledge by a semi-adaptive scoring function. Extensive experiments validate our proposal.

Federated Deep Reinforcement Learning-Based Intelligent Dynamic Services in UAV-Assisted MEC

Unmanned Aerial Vehicles (UAVs)-assisted Multi-access Edge Computing (MEC) has emerged as a promising solution in B5G/6G networks. The high flexibility and seamless connectivity of UAVs make them well-suited for providing enhanced communications coverage and efficient computing support. Particularly in situations where ground facilities may be compromised or communication is unreliable. In this paper, we study joint dynamic service switching and resource allocation for multiple UAVs in MEC network. We consider the heterogeneity of tasks and UAVs and model the dynamic service process of UAVs as a sequential decision problem based on the Markovian decision process. To enable dynamic and intelligent UAV service, we first propose a centralized dynamic service algorithm DDBC based on deep reinforcement learning. However, given the training difficulties of the centralized algorithm, we propose a more promising distributed learning algorithm FLBF, which combines federated learning. We conduct extensive simulations to evaluate the effectiveness and advantages of the proposed algorithms. Our results show that both DDBC and FLBF can improve the model convergence speed by 50%, and reduce the system cost by 12.30% to 35.72% compared to comparative algorithms. Furthermore, simulations indicate that FLBF is well-suited for training models in UAV-assisted MEC networks.

Decentralized Parallel SGD Based on Weight-Balancing for Intelligent IoV

Training machine learning models in a decentralized way has attracted tremendous attention on intelligent Internet of Vehicles (IIoV). However, it is highly dynamic and asymmetric for the connections between vehicles in IIoV due to the mobility of vehicles and the complex communication environment, which poses great challenges on designing efficient distributed learning algorithms. To address this problem, we focus on the basic stochastic gradient descent (SGD) algorithm and propose a decentralized parallel SGD algorithm (DPSGD-WB) for the complex IIoV. The algorithm is based on weight-balancing to overcome the difficulty caused by the dynamic and asymmetric connectivity in IIoV. With rigorous analysis, we show that DPSGD-WB converges on the optimal rate of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$O(1/\sqrt{Kn})$</tex-math> </inline-formula> , where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$n$</tex-math> </inline-formula> is the number of vehicle terminals and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$</tex-math> </inline-formula> is the number of iterations. To the best of our knowledge, our proposed algorithm is the first known decentralized parallel SGD algorithm that can be implemented in asymmetric and dynamic intelligent IoV systems. Finally, extensive experiments demonstrate the efficacy of our algorithm.

Prioritized experience replay based deep distributional reinforcement learning for battery operation in microgrids

Reinforcement Learning (RL) provides a pathway for efficiently utilizing the battery storage in a microgrid. However, traditional value-based RL algorithms used in battery management focus on formulating the policies based on the reward expectation rather than its probability distribution. Hence the scheduling strategy is solely based on the expectation of the rewards rather than the distribution. This paper focuses on scheduling strategy based on probability distribution of the rewards which optimally reflects the uncertainties in the incoming dataset. Furthermore, the prioritized experience replay samples of the training experience are used to enhance the quality of the learning by reducing bias. The results are obtained with different variants of distributional RL algorithms like C51, Quantile Regression Deep Q-Network (QR-DQN), Fully Quantizable Function (FQF), Implicit Quantile Networks (IQN) and rainbow. Moreover, the results are compared with the traditional deep Q-learning algorithm with prioritized experienced replay. The convergence results on the training dataset are further analyzed by varying the action spaces, using randomized experience replay and without including the tariff-based action while enforcing the penalties for violating battery SoC limits. The best trained Q-network is tested with different load and PV profiles to obtain the battery operation and costs. The performance of the distributional RL algorithms is analyzed under different schemes of Time of Use (ToU) tariff. QR-DQN with prioritized experience replay has been found to be the best performing algorithm in terms of convergence on the training dataset, with least fluctuation in validation dataset and battery operations during different tariff regimes during the day.

Exploring Current Practices, Needs, and Barriers for Expanding Distributed Medical Education and Scholarship in Psychiatry: Protocol for an Environmental Scan Using a Formal Information Search Approach and Explanatory Design.

Distributed medical education (DME) offers manifold benefits, such as increased training capacity, enhanced clinical learning, and enhanced rural physician recruitment. Engaged faculty are pivotal to DME's success, necessitating efforts from the academic department to promote integration into scholarly and research activities. Environmental scanning has been used to gather, analyze, and apply information for strategic planning purposes. It helps organizations identify current practices, assess needs and barriers, and respond to emerging risks and opportunities. There are process models and conceptual frameworks developed for environmental scanning in the business and educational sectors. However, the literature lacks methodological direction on how to go about designing and implementing this strategy to guide research and practice in DME, especially in the psychiatry field. This paper presents a protocol for an environmental scanning that aims to understand current practices and identify needs and barriers that must be addressed to facilitate the integration of psychiatrists from the Dalhousie University Faculty of Medicine's distributed education sites in Nova Scotia and New Brunswick into the Department of Psychiatry, contributing for the expansion of DME in both provinces and informing strategic planning and decision-making within the organization. This protocol adopts an innovative approach combining a formal information search and an explanatory design that includes quantitative and qualitative data. About 120 psychiatrists from 8 administrative health zones of both provinces will be invited to complete an anonymous web-based survey with questions about demographics, participants' experience and interest in undergraduate, postgraduate, and continuing medical education, research and scholarly activities, quality improvement, and knowledge translation. Focus group sessions will be conducted with a purposive sample of psychiatrists to collect qualitative data on their perspectives on the expansion of DME. Results are expected within 6 months of data collection and will inform policy options for expanding Dalhousie University's psychiatry residency and fellowship programs using the infrastructure and human resources at distributed learning sites, leveraging opportunities regionally, especially in rural areas. This paper proposes a comprehensive environmental scan procedure adapted from existing approaches. It does this by collecting important characteristics that affect psychiatrists' desire to be involved with research and scholarly activities, which is crucial for the DME expansion. Furthermore, its concordance with the literature facilitates interpretation and comparison. The protocol's new method also fills DME information gaps, allowing one to identify insights and patterns that may shape psychiatric education. This environmental scan's results will answer essential questions about how training programs could involve therapists outside the academic core and make the most of training experiences in semiurban and rural areas. This could help other psychiatry and medical units outside tertiary care establish residency and fellowship programs. DERR1-10.2196/46835.

A Privacy Preserving System for Movie Recommendations Using Federated Learning

Recommender systems have become ubiquitous in the past years. They solve the tyranny of choice problem faced by many users, and are utilized by many online businesses to drive engagement and sales. Besides other criticisms, like creating filter bubbles within social networks, recommender systems are often reproved for collecting considerable amounts of personal data. However, to personalize recommendations, personal information is fundamentally required. A recent distributed learning scheme called federated learning has made it possible to learn from personal user data without its central collection. Consequently, we present a recommender system for movie recommendations, which provides privacy and thus trustworthiness on multiple levels: First and foremost, it is trained using federated learning and thus, by its very nature, privacy-preserving, while still enabling users to benefit from global insights. Furthermore, a novel federated learning scheme, called FedQ, is employed, which not only addresses the problem of non-i.i.d.-ness and small local datasets, but also prevents input data reconstruction attacks by aggregating client updates early. Finally, to reduce the communication overhead, compression is applied, which significantly compresses the exchanged neural network parametrizations to a fraction of their original size. We conjecture that this may also improve data privacy through its lossy quantization stage.

Social Learning and Reputation Management in an Espionage Crisis

The purpose of this qualitative case study comprising 18 thematic interviews was to explore social learning and provide an internal focus on the development path of an international company coping in a crisis. EC is an international engineering and design agency that grew from a small business to a medium-sized enterprise during the crisis era. Since it began operations in 2011, there has been a suspicion of industrial espionage, directed at the founders of the company, including the CEO. Managing internal and external reputations in these circumstances is challenging, as the company builds its legitimacy. The data were content analysed and the results of the study concluded that social learning is crucial to surviving crises. The abductive content analysis and findings of this study are threefold. First, it is suggested that social learning enhances organisational coping in the midst of a crisis. Next, to manage business development and organisational reputation during the crisis, it is necessary to use distributed learning, thinking and doing simultaneously. Thus, the core business and the need for ongoing espionage crime detection create potential harm to the company’s external reputation. Nevertheless, EC’s revenue continuously increased and the trust of clients was maintained.

Geospatial distribution and machine learning algorithms for assessing water quality in surface water bodies of Morocco

Surface waterbodies being primary source of water for human consumption are being investigated for its quality globally. This study evaluated water quality in three rivers (River Nfifikh, Hassar and El Maleh) of Mohammedia prefecture, Morocco in terms of heavy metals occurrence during two seasons of winter and spring. The heavy metals analyzed were cadmium, iron, copper, zinc, and lead. Heavy metal pollution index was derived to quantify water quality and pollution. Hazard quotient and carcinogenic risk were calculated to determine possible health risk. Modelling and prediction were performed using random forest, support vector machine and artificial neural network. The heavy metal concentration was lower in the winter season than in the spring season. Heavy metal pollution index (H.P.I.) was in the range of 1.5–2 during the winter season and 2–3 during the spring season. In the Nfifikh river, Cd2+ and Fe were the main polluting heavy metal. H.Q. was < 1 in all three rivers, which signified no adverse health effect from exposure to heavy metals. However, carcinogenic risk assessment revealed that 1 in every 100 people was susceptible to cancer during the life span of 70 years. Based on the control point reference, it was found that Mohammedia prefecture as river water was already contaminated before it entered the prefecture boundary. This was again validated with the water lagoon Douar El Marja which is located near the industrial zones of Mohammedia prefecture. Future studies are required to investigate pollution of rivers prior to their entry in Mohammedia prefecture to identify potential source and adopt mitigation measures accordingly.

Estimates on learning rates for multi-penalty distribution regression

This paper is concerned with functional learning by utilizing two-stage sampled distribution regression. We study a multi-penalty regularization algorithm for distribution regression in the framework of learning theory. The algorithm aims at regressing to real-valued outputs from probability measures. The theoretical analysis of distribution regression is far from maturity and quite challenging since only second-stage samples are observable in practical settings. In our algorithm, to transform information of distribution samples, we embed the distributions to a reproducing kernel Hilbert space HK associated with Mercer kernel K via mean embedding technique. One of the primary contributions of this work is the introduction of a novel multi-penalty regularization algorithm, which is able to capture more potential features of distribution regression. Optimal learning rates of the algorithm are obtained under mild conditions. The work also derives learning rates for distribution regression in the hard learning scenario fρ∉HK, which has not been explored in the existing literature. Moreover, we propose a new distribution-regression-based distributed learning algorithm to face large-scale data or information challenges arising from distribution data. The optimal learning rates are derived for the distributed learning algorithm. By providing new algorithms and showing their learning rates, the work improves the existing literature in various aspects.

Universal adversarial backdoor attacks to fool vertical federated learning

Vertical federated learning (VFL) is a privacy-preserving distribution learning paradigm that enables participants, owning different features of the same sample space to train a machine learning model collaboratively while retaining their data locally. This paradigm facilitates improved efficiency and security for participants such as financial or medical fields, making VFL an essential component of data-driven Artificial Intelligence systems. Nevertheless, the partitioned structure of VFL can be exploited by adversaries to inject a backdoor, enabling them to manipulate the VFL predictions. In this paper, we aim to investigate the vulnerability of VFL in the context of binary classification tasks. To this end, we define a threat model for backdoor attacks in VFL and introduce a universal adversarial backdoor (UAB) attack to poison the predictions of VFL. The UAB attack, consisting of universal trigger generation and clean-label backdoor injection, is incorporated during the VFL training at specific iterations. This is achieved by alternately optimizing VFL sub-problems' universal trigger and model parameters. Our work distinguishes itself from existing studies on designing backdoor attacks for VFL, as those require the knowledge of auxiliary information that is not accessible within the split VFL architecture. In contrast, our approach does not require additional data to execute the attack. On the real-world datasets, our approach surpasses existing state-of-the-art methods, achieving up to 100% backdoor task performance while maintaining the main task performance. Our results in this paper make a major advance in revealing the hidden backdoor risks of VFL, hence paving the way for the future development of secure VFL. Our results in this paper make a major advance in revealing the hidden backdoor risks of VFL, hence paving the way for the future development of secure VFL applications such as finance.

Alleviating sample imbalance in water quality assessment using the VAE–WGAN–GP model

Abstract Water resources are essential for sustaining human life and promoting sustainable development. However, rapid urbanization and industrialization have resulted in a decline in freshwater availability. Effective prevention and control of water pollution are essential for ecological balance and human well-being. Water quality assessment is crucial for monitoring and managing water resources. Existing machine learning-based assessment methods tend to classify the results into the majority class, leading to inaccuracies in the outcomes due to the prevalent issue of imbalanced class sample distribution in practical scenarios. To tackle the issue, we propose a novel approach that utilizes the VAE–WGAN–GP model. The VAE–WGAN–GP model combines the encoding and decoding mechanisms of VAE with the adversarial learning of GAN. It generates synthetic samples that closely resemble real samples, effectively compensating data of the scarcity category in water quality evaluation. Our contributions include (1) introducing a deep generative model to alleviate the issue of imbalanced category samples in water quality assessment, (2) demonstrating the faster convergence speed and improved potential distribution learning ability of the proposed VAE–WGAN–GP model, (3) introducing the compensation degree concept and conducting comprehensive compensation experiments, resulting in a 9.7% increase in the accuracy of water quality assessment for multi-classification imbalance samples.

Distributed Computing and Inference for Big Data

Data are distributed across different sites due to computing facility limitations or data privacy considerations. Conventional centralized methods—those in which all datasets are stored and processed in a central computing facility—are not applicable in practice. Therefore, it has become necessary to develop distributed learning approaches that have good inference or predictive accuracy while remaining free of individual data or obeying policies and regulations to protect privacy. In this article, we introduce the basic idea of distributed learning and conduct a selected review on various distributed learning methods, which are categorized by their statistical accuracy, computational efficiency, heterogeneity, and privacy. This categorization can help evaluate newly proposed methods from different aspects. Moreover, we provide up-to-date descriptions of the existing theoretical results that cover statistical equivalency and computational efficiency under different statistical learning frameworks. Finally, we provide existing software implementations and benchmark datasets, and we discuss future research opportunities.

Modeling annotator preference and stochastic annotation error for medical image segmentation

Manual annotation of medical images is highly subjective, leading to inevitable annotation biases. Deep learning models may surpass human performance on a variety of tasks, but they may also mimic or amplify these biases. Although we can have multiple annotators and fuse their annotations to reduce stochastic errors, we cannot use this strategy to handle the bias caused by annotators’ preferences. In this paper, we highlight the issue of annotator-related biases on medical image segmentation tasks, and propose a Preference-involved Annotation Distribution Learning (PADL) framework to address it from the perspective of modeling an annotator’s preference and stochastic errors so as to produce not only a meta segmentation but also the annotator-specific segmentation. Under this framework, a stochastic error modeling (SEM) module estimates the meta segmentation map and average stochastic error map, and a series of human preference modeling (HPM) modules estimate each annotator’s segmentation and the corresponding stochastic error. We evaluated our PADL framework on two medical image benchmarks with different imaging modalities, which have been annotated by multiple medical professionals, and achieved promising performance on all five medical image segmentation tasks. Code is available at https://github.com/Merrical/PADL.

FedER: Federated Learning through Experience Replay and privacy-preserving data synthesis

In the medical field, multi-center collaborations are often sought to yield more generalizable findings by leveraging the heterogeneity of patient and clinical data. However, recent privacy regulations hinder the possibility to share data, and consequently, to come up with machine learning-based solutions that support diagnosis and prognosis. Federated learning (FL) aims at sidestepping this limitation by bringing AI-based solutions to data owners and only sharing local AI models, or parts thereof, that need then to be aggregated. However, most of the existing federated learning solutions are still at their infancy and show several shortcomings, from the lack of a reliable and effective aggregation scheme able to retain the knowledge learned locally to weak privacy preservation as real data may be reconstructed from model updates. Furthermore, the majority of these approaches, especially those dealing with medical data, relies on a centralized distributed learning strategy that poses robustness, scalability and trust issues. In this paper we present a federated learning strategy, FedER, that, exploiting experience replay and generative adversarial concepts, effectively integrates features from local nodes, providing models able to generalize across multiple datasets while maintaining privacy. FedER is tested on two tasks — tuberculosis and melanoma classification — using multiple datasets in order to simulate realistic non-i.i.d. medical data scenarios. Results show that our approach achieves performance comparable to standard (non-federated) learning and significantly outperforms state-of-the-art federated methods. Remarkably, we also observe that FedER enables any node model to be used as a global federation model. Indeed, the experience replay strategy with privacy-preserving synthetic data allows all node models to converge to reach the same optimum without the need of a single shared model. Code is available at https://github.com/perceivelab/FedER.

GE-DDRL: Graph Embedding and Deep Distributional Reinforcement Learning for Reliable Shortest Path: A Universal and Scale Free Solution

This paper studies the reliable shortest path (RSP) problem in stochastic transportation networks. State-of-the-art RSP solutions usually target one specific RSP problem; moreover, the corresponding algorithm’s computational complexity scales at least linearly with the size of the underlying transportation network. While in this paper, we propose a graph embedding and deep distributional reinforcement learning (GE-DDRL) method, which serves as a universal and scale-free solution to the RSP problem. GE-DDRL uses deep distributional reinforcement learning (DDRL) to estimate the full travel-time distribution of a given routing policy, and improves the given routing policy with the generalized policy iteration (GPI) scheme. Further, in order to achieve the generalization ability to new destination nodes, we employ one of the canonical graph embedding techniques (Skip-Gram) to compress the nodes’ representation into <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$d$</tex-math> </inline-formula> -dimensional real-valued vectors. With the properly compressed node features, GE-DDRL is able to generalize its estimation of the routing policy’s travel-time distribution to untrained destination nodes, and hence achieve the ‘all-to-all’ navigation functionality. To the best of our knowledge, GE-DDRL serves as the first RSP planner, which applies simultaneously to almost all RSP objectives and in the meanwhile, is scale free with the size of the transportation network in terms of the online decision-making time and memory complexity. Experimental results and comparisons with state of the arts show the efficacy and efficiency of GE-DDRL in a range of transportation networks.

Intelligent Electric Vehicle Charging Scheduling in Transportation-Energy Nexus with Distributional Reinforcement Learning

Intelligent Electric Vehicle Charging Scheduling in Transportation-Energy Nexus with Distributional Reinforcement Learning

Portfolio selection under non-gaussianity and systemic risk: A machine learning based forecasting approach

The Sharpe-ratio-maximizing portfolio becomes questionable under non-Gaussian returns, and it rules out, by construction, systemic risk, which can negatively affect its out-of-sample performance. In the present work, we develop a new performance ratio that simultaneously addresses these two problems when building optimal portfolios. To robustify the portfolio optimization and better represent extreme market scenarios, we simulate a large number of returns via a Monte Carlo method. This is done by obtaining probabilistic return forecasts through a distributional machine learning approach in a big data setting and then combining them with a fitted copula to generate return scenarios. Based on a large-scale comparative analysis conducted on the US market, the backtesting results demonstrate the superiority of our proposed portfolio selection approach against several popular benchmark strategies in terms of both profitability and minimizing systemic risk. This outperformance is robust to the inclusion of transaction costs.