Machine Learning (cs.LG)

Abstract: Transfer learning leverages feature representations of deep neural networks (DNNs) pretrained on source tasks with rich data to empower effective finetuning on downstream tasks. However, the pretrained models are often prohibitively large for delivering generalizable representations, which limits their deployment on edge devices with constrained resources. To close this gap, we propose a new transfer learning pipeline, which leverages our finding that robust tickets can transfer better, i.e., subnetworks drawn with properly induced adversarial robustness can win better transferability over vanilla lottery ticket subnetworks. Extensive experiments and ablation studies validate that our proposed transfer learning pipeline can achieve enhanced accuracy-sparsity trade-offs across both diverse downstream tasks and sparsity patterns, further enriching the lottery ticket hypothesis.

Abstract: This paper presents a local energy distribution based hyperparameter determination for stochastic simulated annealing (SSA). SSA is capable of solving combinatorial optimization problems faster than typical simulated annealing (SA), but requires a time-consuming hyperparameter search. The proposed method determines hyperparameters based on the local energy distributions of spins (probabilistic bits). The spin is a basic computing element of SSA and is graphically connected to other spins with its weights. The distribution of the local energy can be estimated based on the central limit theorem (CLT). The CLT-based normal distribution is used to determine the hyperparameters, which reduces the time complexity for hyperparameter search from O(n^3) of the conventional method to O(1). The performance of SSA with the determined hyperparameters is evaluated on the Gset and K2000 benchmarks for maximum-cut problems. The results show that the proposed method achieves mean cut values of approximately 98% of the best-known cut values.

Abstract: Machine learning (ML) models have been quite successful in predicting outcomes in many applications. However, in some cases, domain experts might have a judgment about the expected outcome that might conflict with the prediction of ML models. One main reason for this is that the training data might not be totally representative of the population. In this paper, we present a novel framework that aims at leveraging experts' judgment to mitigate the conflict. The underlying idea behind our framework is that we first determine, using a generative adversarial network, the degree of representation of an unlabeled data point in the training data. Then, based on such degree, we correct the \textcolor{black}{machine learning} model's prediction by incorporating the experts' judgment into it, where the higher that aforementioned degree of representation, the less the weight we put on the expert intuition that we add to our corrected output, and vice-versa. We perform multiple numerical experiments on synthetic data as well as two real-world case studies (one from the IT services industry and the other from the financial industry). All results show the effectiveness of our framework; it yields much higher closeness to the experts' judgment with minimal sacrifice in the prediction accuracy, when compared to multiple baseline methods. We also develop a new evaluation metric that combines prediction accuracy with the closeness to experts' judgment. Our framework yields statistically significant results when evaluated on that metric.

Abstract: Underwater acoustic recognition for ship-radiated signals has high practical application value due to the ability to recognize non-line-of-sight targets. However, due to the difficulty of data acquisition, the collected signals are scarce in quantity and mainly composed of mechanical periodic noise. According to the experiments, we observe that the repeatability of periodic signals leads to a double-descent phenomenon, which indicates a significant local bias toward repeated samples. To address this issue, we propose a strategy based on cross-entropy to prune excessively similar segments in training data. Furthermore, to compensate for the reduction of training data, we generate noisy samples and apply smoothness-inducing regularization based on KL divergence to mitigate overfitting. Experiments show that our proposed data pruning and regularization strategy can bring stable benefits and our framework significantly outperforms the state-of-the-art in low-resource scenarios.

Abstract: Large data-driven image models are extensively used to support creative and artistic work. Under the currently predominant distribution-fitting paradigm, a dataset is treated as ground truth to be approximated as closely as possible. Yet, many creative applications demand a diverse range of output, and creators often strive to actively diverge from a given data distribution. We argue that an adjustment of modelling objectives, from pure mode coverage towards mode balancing, is necessary to accommodate the goal of higher output diversity. We present diversity weights, a training scheme that increases a model's output diversity by balancing the modes in the training dataset. First experiments in a controlled setting demonstrate the potential of our method. We conclude by contextualising our contribution to diversity within the wider debate on bias, fairness and representation in generative machine learning.

Abstract: An essential problem in causal inference is estimating causal effects from observational data. The problem becomes more challenging with the presence of unobserved confounders. When there are unobserved confounders, the commonly used back-door adjustment is not applicable. Although the instrumental variable (IV) methods can deal with unobserved confounders, they all assume that the treatment directly affects the outcome, and there is no mediator between the treatment and the outcome. This paper aims to use the front-door criterion to address the challenging problem with the presence of unobserved confounders and mediators. In practice, it is often difficult to identify the set of variables used for front-door adjustment from data. By leveraging the ability of deep generative models in representation learning, we propose FDVAE to learn the representation of a Front-Door adjustment set with a Variational AutoEncoder, instead of trying to search for a set of variables for front-door adjustment. Extensive experiments on synthetic datasets validate the effectiveness of FDVAE and its superiority over existing methods. The experiments also show that the performance of FDVAE is not sensitive to the causal strength of unobserved confounders and is feasible in the case of dimensionality mismatch between learned representations and the ground truth. We further apply the method to three real-world datasets to demonstrate its potential applications.

Abstract: Many score-based active learning methods have been successfully applied to graph-structured data, aiming to reduce the number of labels and achieve better performance of graph neural networks based on predefined score functions. However, these algorithms struggle to learn policy distributions that are proportional to rewards and have limited exploration capabilities. In this paper, we innovatively formulate the graph active learning problem as a generative process, named GFlowGNN, which generates various samples through sequential actions with probabilities precisely proportional to a predefined reward function. Furthermore, we propose the concept of flow nodes and flow features to efficiently model graphs as flows based on generative flow networks, where the policy network is trained with specially designed rewards. Extensive experiments on real datasets show that the proposed approach has good exploration capability and transferability, outperforming various state-of-the-art methods.

Abstract: The real-world implementation of federated learning is complex and requires research and development actions at the crossroad between different domains ranging from data science, to software programming, networking, and security. While today several FL libraries are proposed to data scientists and users, most of these frameworks are not designed to find seamless application in medical use-cases, due to the specific challenges and requirements of working with medical data and hospital infrastructures. Moreover, governance, design principles, and security assumptions of these frameworks are generally not clearly illustrated, thus preventing the adoption in sensitive applications. Motivated by the current technological landscape of FL in healthcare, in this document we present Fed-BioMed: a research and development initiative aiming at translating federated learning (FL) into real-world medical research applications. We describe our design space, targeted users, domain constraints, and how these factors affect our current and future software architecture.

Abstract: Node representation learning in a network is an important machine learning technique for encoding relational information in a continuous vector space while preserving the inherent properties and structures of the network. Recently, \textit{unsupervised} node embedding methods such as DeepWalk \citep{deepwalk}, LINE \citep{line}, struc2vec \citep{struc2vec}, PTE \citep{pte}, UserItem2vec \citep{wu2020multi}, and RWJBG \citep{li2021random} have emerged from the Skip-gram model \citep{word2vec} and perform better performance in several downstream tasks such as node classification and link prediction than the existing relational models. However, providing post-hoc explanations of Skip-gram-based embeddings remains a challenging problem because of the lack of explanation methods and theoretical studies applicable for embeddings. In this paper, we first show that global explanations to the Skip-gram-based embeddings can be found by computing \textit{bridgeness} under a spectral cluster-aware local perturbation. Moreover, a novel gradient-based explanation method, which we call GRAPH-wGD, is proposed that allows the top-$q$ global explanations about learned graph embedding vectors more efficiently. Experiments show that the ranking of nodes by scores using GRAPH-wGD is highly correlated with true \textit{bridgeness} scores. We also observe that the top-$q$ node-level explanations selected by GRAPH-wGD have higher importance scores and produce more changes in class label prediction when perturbed, compared with the nodes selected by recent alternatives, using five real-world graphs.

Abstract: Continual learning enables the incremental training of machine learning models on non-stationary data streams.While academic interest in the topic is high, there is little indication of the use of state-of-the-art continual learning algorithms in practical machine learning deployment. This paper presents Renate, a continual learning library designed to build real-world updating pipelines for PyTorch models. We discuss requirements for the use of continual learning algorithms in practice, from which we derive design principles for Renate. We give a high-level description of the library components and interfaces. Finally, we showcase the strengths of the library by presenting experimental results. Renate may be found at https://github.com/awslabs/renate.

Abstract: Customers' load profiles are critical resources to support data analytics applications in modern power systems. However, there are usually insufficient historical load profiles for data analysis, due to the collection cost and data privacy issues. To address such data shortage problems, load profiles synthesis is an effective technique that provides synthetic training data for customers to build high-performance data-driven models. Nonetheless, it is still challenging to synthesize high-quality load profiles for each customer using generation models trained by the respective customer's data owing to the high heterogeneity of customer load. In this paper, we propose a novel customized load profiles synthesis method based on conditional diffusion models for heterogeneous customers. Specifically, we first convert the customized synthesis into a conditional data generation issue. We then extend traditional diffusion models to conditional diffusion models to realize conditional data generation, which can synthesize exclusive load profiles for each customer according to the customer's load characteristics and application demands. In addition, to implement conditional diffusion models, we design a noise estimation model with stacked residual layers, which improves the generation performance by using skip connections. The attention mechanism is also utilized to better extract the complex temporal dependency of load profiles. Finally, numerical case studies based on a public dataset are conducted to validate the effectiveness and superiority of the proposed method.

Abstract: In this paper, we propose ComGAN(ComparativeGAN) which allows the generator in GANs to refer to the semantics of comparative samples(e.g. real data) by comparison. ComGAN generalizes relativistic GANs by using arbitrary architecture and mostly outperforms relativistic GANs in simple input-concatenation architecture. To train the discriminator in ComGAN, we also propose equality regularization, which fits the discriminator to a neutral label for equally real or fake samples. Equality regularization highly boosts the performance of ComGAN including WGAN while being exceptionally simple compared to existing regularizations. Finally, we generalize comparative samples fixed to real data in relativistic GANs toward fake data and show that such objectives are sound in both theory and practice. Our experiments demonstrate superior performances of ComGAN and equality regularization, achieving the best FIDs in 7 out of 8 cases of different losses and data against ordinary GANs and relativistic GANs.

Abstract: This paper presents FedPIDAvg, the winning submission to the Federated Tumor Segmentation Challenge 2022 (FETS22). Inspired by FedCostWAvg, our winning contribution to FETS21, we contribute an improved aggregation strategy for federated and collaborative learning. FedCostWAvg is a weighted averaging method that not only considers the number of training samples of each cluster but also the size of the drop of the respective cost function in the last federated round. This can be interpreted as the derivative part of a PID controller (proportional-integral-derivative controller). In FedPIDAvg, we further add the missing integral term. Another key challenge was the vastly varying size of data samples per center. We addressed this by modeling the data center sizes as following a Poisson distribution and choosing the training iterations per center accordingly. Our method outperformed all other submissions.

Abstract: Variational auto-encoders (VAEs) are one of the most popular approaches to deep generative modeling. Despite their success, images generated by VAEs are known to suffer from blurriness, due to a highly unrealistic modeling assumption that the conditional data distribution $ p(\textbf{x} | \textbf{z})$ can be approximated as an isotropic Gaussian. In this work we introduce a principled approach to modeling the conditional data distribution $p(\textbf{x} | \textbf{z})$ by incorporating a diffusion model. We show that it is possible to create a VAE-like deep latent variable model without making the Gaussian assumption on $ p(\textbf{x} | \textbf{z}) $ or even training a decoder network. A trained encoder and an unconditional diffusion model can be combined via Bayes' rule for score functions to obtain an expressive model for $ p(\textbf{x} | \textbf{z}) $. Our approach avoids making strong assumptions on the parametric form of $ p(\textbf{x} | \textbf{z}) $, and thus allows to significantly improve the performance of VAEs.

Abstract: This paper investigates policy resilience to training-environment poisoning attacks on reinforcement learning (RL) policies, with the goal of recovering the deployment performance of a poisoned RL policy. Due to the fact that the policy resilience is an add-on concern to RL algorithms, it should be resource-efficient, time-conserving, and widely applicable without compromising the performance of RL algorithms. This paper proposes such a policy-resilience mechanism based on an idea of knowledge sharing. We summarize the policy resilience as three stages: preparation, diagnosis, recovery. Specifically, we design the mechanism as a federated architecture coupled with a meta-learning manner, pursuing an efficient extraction and sharing of the environment knowledge. With the shared knowledge, a poisoned agent can quickly identify the deployment condition and accordingly recover its policy performance. We empirically evaluate the resilience mechanism for both model-based and model-free RL algorithms, showing its effectiveness and efficiency in restoring the deployment performance of a poisoned policy.

Abstract: Self-supervised learning, dubbed the dark matter of intelligence, is a promising path to advance machine learning. Yet, much like cooking, training SSL methods is a delicate art with a high barrier to entry. While many components are familiar, successfully training a SSL method involves a dizzying set of choices from the pretext tasks to training hyper-parameters. Our goal is to lower the barrier to entry into SSL research by laying the foundations and latest SSL recipes in the style of a cookbook. We hope to empower the curious researcher to navigate the terrain of methods, understand the role of the various knobs, and gain the know-how required to explore how delicious SSL can be.

Abstract: Heterogeneous graph neural networks (HGNNs) as an emerging technique have shown superior capacity of dealing with heterogeneous information network (HIN). However, most HGNNs follow a semi-supervised learning manner, which notably limits their wide use in reality since labels are usually scarce in real applications. Recently, contrastive learning, a self-supervised method, becomes one of the most exciting learning paradigms and shows great potential when there are no labels. In this paper, we study the problem of self-supervised HGNNs and propose a novel co-contrastive learning mechanism for HGNNs, named HeCo. Different from traditional contrastive learning which only focuses on contrasting positive and negative samples, HeCo employs cross-view contrastive mechanism. Specifically, two views of a HIN (network schema and meta-path views) are proposed to learn node embeddings, so as to capture both of local and high-order structures simultaneously. Then the cross-view contrastive learning, as well as a view mask mechanism, is proposed, which is able to extract the positive and negative embeddings from two views. This enables the two views to collaboratively supervise each other and finally learn high-level node embeddings. Moreover, to further boost the performance of HeCo, two additional methods are designed to generate harder negative samples with high quality. Besides the invariant factors, view-specific factors complementally provide the diverse structure information between different nodes, which also should be contained into the final embeddings. Therefore, we need to further explore each view independently and propose a modified model, called HeCo++. Specifically, HeCo++ conducts hierarchical contrastive learning, including cross-view and intra-view contrasts, which aims to enhance the mining of respective structures.

Abstract: We build universal approximators of continuous maps between arbitrary Polish metric spaces $\mathcal{X}$ and $\mathcal{Y}$ using universal approximators between Euclidean spaces as building blocks. Earlier results assume that the output space $\mathcal{Y}$ is a topological vector space. We overcome this limitation by "randomization": our approximators output discrete probability measures over $\mathcal{Y}$. When $\mathcal{X}$ and $\mathcal{Y}$ are Polish without additional structure, we prove very general qualitative guarantees; when they have suitable combinatorial structure, we prove quantitative guarantees for H\"older-like maps, including maps between finite graphs, solution operators to rough differential equations between certain Carnot groups, and continuous non-linear operators between Banach spaces arising in inverse problems. In particular, we show that the required number of Dirac measures is determined by the combinatorial structure of $\mathcal{X}$ and $\mathcal{Y}$. For barycentric $\mathcal{Y}$, including Banach spaces, $\mathbb{R}$-trees, Hadamard manifolds, or Wasserstein spaces on Polish metric spaces, our approximators reduce to $\mathcal{Y}$-valued functions. When the Euclidean approximators are neural networks, our constructions generalize transformer networks, providing a new probabilistic viewpoint of geometric deep learning.

Abstract: Population censuses are vital to public policy decision-making. They provide insight into human resources, demography, culture, and economic structure at local, regional, and national levels. However, such surveys are very expensive (especially for low and middle-income countries with high populations, such as India), time-consuming, and may also raise privacy concerns, depending upon the kinds of data collected. In light of these issues, we introduce SynthPop++, a novel hybrid framework, which can combine data from multiple real-world surveys (with different, partially overlapping sets of attributes) to produce a real-scale synthetic population of humans. Critically, our population maintains family structures comprising individuals with demographic, socioeconomic, health, and geolocation attributes: this means that our ``fake'' people live in realistic locations, have realistic families, etc. Such data can be used for a variety of purposes: we explore one such use case, Agent-based modelling of infectious disease in India. To gauge the quality of our synthetic population, we use both machine learning and statistical metrics. Our experimental results show that synthetic population can realistically simulate the population for various administrative units of India, producing real-scale, detailed data at the desired level of zoom -- from cities, to districts, to states, eventually combining to form a country-scale synthetic population.