Machine Learning (stat.ML)

Abstract: Estimating average causal effects is a common practice to test new treatments. However, the average effect ''masks'' important individual characteristics in the counterfactual distribution, which may lead to safety, fairness, and ethical concerns. This issue is exacerbated in the temporal setting, where the treatment is sequential and time-varying, leading to an intricate influence on the counterfactual distribution. In this paper, we propose a novel conditional generative modeling approach to capture the whole counterfactual distribution, allowing efficient inference on certain statistics of the counterfactual distribution. This makes the proposed approach particularly suitable for healthcare and public policy making. Our generative modeling approach carefully tackles the distribution mismatch in the observed data and the targeted counterfactual distribution via a marginal structural model. Our method outperforms state-of-the-art baselines on both synthetic and real data.

Abstract: The research explores the influence of preschool attendance (one year before full-time school) on the development of children during their first year of school. Using data collected by the Australian Early Development Census, the findings show that areas with high proportions of preschool attendance tended to have lower proportions of children with at least one developmental vulnerability. Developmental vulnerablities include not being able to cope with the school day (tired, hungry, low energy), unable to get along with others or aggressive behaviour, trouble with reading/writing or numbers. These findings, of course, vary by region. Using Data Analysis and Machine Learning, the researchers were able to identify three distinct clusters within Queensland, each characterised by different socio-demographic variables influencing the relationship between preschool attendance and developmental vulnerability. These analyses contribute to understanding regions with high vulnerability and the potential need for tailored policies or investments

Abstract: Diffusion models (DMs) are widely used for generating high-quality image datasets. However, since they operate directly in the high-dimensional pixel space, optimization of DMs is computationally expensive, requiring long training times. This contributes to large amounts of noise being injected into the differentially private learning process, due to the composability property of differential privacy. To address this challenge, we propose training Latent Diffusion Models (LDMs) with differential privacy. LDMs use powerful pre-trained autoencoders to reduce the high-dimensional pixel space to a much lower-dimensional latent space, making training DMs more efficient and fast. Unlike [Ghalebikesabi et al., 2023] that pre-trains DMs with public data then fine-tunes them with private data, we fine-tune only the attention modules of LDMs at varying layers with privacy-sensitive data, reducing the number of trainable parameters by approximately 96% compared to fine-tuning the entire DM. We test our algorithm on several public-private data pairs, such as ImageNet as public data and CIFAR10 and CelebA as private data, and SVHN as public data and MNIST as private data. Our approach provides a promising direction for training more powerful, yet training-efficient differentially private DMs that can produce high-quality synthetic images.

Abstract: We consider contextual bandit problems with knapsacks [CBwK], a problem where at each round, a scalar reward is obtained and vector-valued costs are suffered. The learner aims to maximize the cumulative rewards while ensuring that the cumulative costs are lower than some predetermined cost constraints. We assume that contexts come from a continuous set, that costs can be signed, and that the expected reward and cost functions, while unknown, may be uniformly estimated -- a typical assumption in the literature. In this setting, total cost constraints had so far to be at least of order $T^{3/4}$, where $T$ is the number of rounds, and were even typically assumed to depend linearly on $T$. We are however motivated to use CBwK to impose a fairness constraint of equalized average costs between groups: the budget associated with the corresponding cost constraints should be as close as possible to the natural deviations, of order $\sqrt{T}$. To that end, we introduce a dual strategy based on projected-gradient-descent updates, that is able to deal with total-cost constraints of the order of $\sqrt{T}$ up to poly-logarithmic terms. This strategy is more direct and simpler than existing strategies in the literature. It relies on a careful, adaptive, tuning of the step size.

Abstract: The approximation properties of infinitely wide shallow neural networks heavily depend on the choice of the activation function. To understand this influence, we study embeddings between Barron spaces with different activation functions. These embeddings are proven by providing push-forward maps on the measures $\mu$ used to represent functions $f$. An activation function of particular interest is the rectified power unit ($\operatorname{RePU}$) given by $\operatorname{RePU}_s(x)=\max(0,x)^s$. For many commonly used activation functions, the well-known Taylor remainder theorem can be used to construct a push-forward map, which allows us to prove the embedding of the associated Barron space into a Barron space with a $\operatorname{RePU}$ as activation function. Moreover, the Barron spaces associated with the $\operatorname{RePU}_s$ have a hierarchical structure similar to the Sobolev spaces $H^m$.

Abstract: Simulation-based inference (SBI) methods such as approximate Bayesian computation (ABC), synthetic likelihood, and neural posterior estimation (NPE) rely on simulating statistics to infer parameters of intractable likelihood models. However, such methods are known to yield untrustworthy and misleading inference outcomes under model misspecification, thus hindering their widespread applicability. In this work, we propose the first general approach to handle model misspecification that works across different classes of SBI methods. Leveraging the fact that the choice of statistics determines the degree of misspecification in SBI, we introduce a regularized loss function that penalises those statistics that increase the mismatch between the data and the model. Taking NPE and ABC as use cases, we demonstrate the superior performance of our method on high-dimensional time-series models that are artificially misspecified. We also apply our method to real data from the field of radio propagation where the model is known to be misspecified. We show empirically that the method yields robust inference in misspecified scenarios, whilst still being accurate when the model is well-specified.

Abstract: Identifiability of latent variable models has recently gained interest in terms of its applications to interpretability or out of distribution generalisation. In this work, we study identifiability of Markov Switching Models as a first step towards extending recent results to sequential latent variable models. We present identifiability conditions within first-order Markov dependency structures, and parametrise the transition distribution via non-linear Gaussians. Our experiments showcase the applicability of our approach for regime-dependent causal discovery and high-dimensional time series segmentation.

Abstract: We study the problem of approximate sampling from non-log-concave distributions, e.g., Gaussian mixtures, which is often challenging even in low dimensions due to their multimodality. We focus on performing this task via Markov chain Monte Carlo (MCMC) methods derived from discretizations of the overdamped Langevin diffusions, which are commonly known as Langevin Monte Carlo algorithms. Furthermore, we are also interested in two nonsmooth cases for which a large class of proximal MCMC methods have been developed: (i) a nonsmooth prior is considered with a Gaussian mixture likelihood; (ii) a Laplacian mixture distribution. Such nonsmooth and non-log-concave sampling tasks arise from a wide range of applications to Bayesian inference and imaging inverse problems such as image deconvolution. We perform numerical simulations to compare the performance of most commonly used Langevin Monte Carlo algorithms.

Abstract: A core principle in statistical learning is that smoothness of target functions allows to break the curse of dimensionality. However, learning a smooth function through Taylor expansions requires enough samples close to one another to get meaningful estimate of high-order derivatives, which seems hard in machine learning problems where the ratio between number of data and input dimension is relatively small. Should we really hope to break the curse of dimensionality based on Taylor expansion estimation? What happens if Taylor expansions are replaced by Fourier or wavelet expansions? By deriving a new lower bound on the generalization error, this paper investigates the role of constants and transitory regimes which are usually not depicted beyond classical learning theory statements while that play a dominant role in practice.

Abstract: We propose EB-TC$\varepsilon$, a novel sampling rule for $\varepsilon$-best arm identification in stochastic bandits. It is the first instance of Top Two algorithm analyzed for approximate best arm identification. EB-TC$\varepsilon$ is an *anytime* sampling rule that can therefore be employed without modification for fixed confidence or fixed budget identification (without prior knowledge of the budget). We provide three types of theoretical guarantees for EB-TC$\varepsilon$. First, we prove bounds on its expected sample complexity in the fixed confidence setting, notably showing its asymptotic optimality in combination with an adaptive tuning of its exploration parameter. We complement these findings with upper bounds on its probability of error at any time and for any error parameter, which further yield upper bounds on its simple regret at any time. Finally, we show through numerical simulations that EB-TC$\varepsilon$ performs favorably compared to existing algorithms, in different settings.

Abstract: We propose a new class of generative models that naturally handle data of varying dimensionality by jointly modeling the state and dimension of each datapoint. The generative process is formulated as a jump diffusion process that makes jumps between different dimensional spaces. We first define a dimension destroying forward noising process, before deriving the dimension creating time-reversed generative process along with a novel evidence lower bound training objective for learning to approximate it. Simulating our learned approximation to the time-reversed generative process then provides an effective way of sampling data of varying dimensionality by jointly generating state values and dimensions. We demonstrate our approach on molecular and video datasets of varying dimensionality, reporting better compatibility with test-time diffusion guidance imputation tasks and improved interpolation capabilities versus fixed dimensional models that generate state values and dimensions separately.

Abstract: Neural SDEs are continuous-time generative models for sequential data. State-of-the-art performance for irregular time series generation has been previously obtained by training these models adversarially as GANs. However, as typical for GAN architectures, training is notoriously unstable, often suffers from mode collapse, and requires specialised techniques such as weight clipping and gradient penalty to mitigate these issues. In this paper, we introduce a novel class of scoring rules on pathspace based on signature kernels and use them as objective for training Neural SDEs non-adversarially. By showing strict properness of such kernel scores and consistency of the corresponding estimators, we provide existence and uniqueness guarantees for the minimiser. With this formulation, evaluating the generator-discriminator pair amounts to solving a system of linear path-dependent PDEs which allows for memory-efficient adjoint-based backpropagation. Moreover, because the proposed kernel scores are well-defined for paths with values in infinite dimensional spaces of functions, our framework can be easily extended to generate spatiotemporal data. Our procedure permits conditioning on a rich variety of market conditions and significantly outperforms alternative ways of training Neural SDEs on a variety of tasks including the simulation of rough volatility models, the conditional probabilistic forecasts of real-world forex pairs where the conditioning variable is an observed past trajectory, and the mesh-free generation of limit order book dynamics.

Abstract: Quantifying the uncertainty of quantities of interest (QoIs) from physical systems is a primary objective in model validation. However, achieving this goal entails balancing the need for computational efficiency with the requirement for numerical accuracy. To address this trade-off, we propose a novel bi-fidelity formulation of variational auto-encoders (BF-VAE) designed to estimate the uncertainty associated with a QoI from low-fidelity (LF) and high-fidelity (HF) samples of the QoI. This model allows for the approximation of the statistics of the HF QoI by leveraging information derived from its LF counterpart. Specifically, we design a bi-fidelity auto-regressive model in the latent space that is integrated within the VAE's probabilistic encoder-decoder structure. An effective algorithm is proposed to maximize the variational lower bound of the HF log-likelihood in the presence of limited HF data, resulting in the synthesis of HF realizations with a reduced computational cost. Additionally, we introduce the concept of the bi-fidelity information bottleneck (BF-IB) to provide an information-theoretic interpretation of the proposed BF-VAE model. Our numerical results demonstrate that BF-VAE leads to considerably improved accuracy, as compared to a VAE trained using only HF data when limited HF data is available.

Abstract: Deep neural networks (DNNs) trained to minimize a loss term plus the sum of squared weights via gradient descent corresponds to the common approach of training with weight decay. This paper provides new insights into this common learning framework. We characterize the kinds of functions learned by training with weight decay for multi-output (vector-valued) ReLU neural networks. This extends previous characterizations that were limited to single-output (scalar-valued) networks. This characterization requires the definition of a new class of neural function spaces that we call vector-valued variation (VV) spaces. We prove that neural networks (NNs) are optimal solutions to learning problems posed over VV spaces via a novel representer theorem. This new representer theorem shows that solutions to these learning problems exist as vector-valued neural networks with widths bounded in terms of the number of training data. Next, via a novel connection to the multi-task lasso problem, we derive new and tighter bounds on the widths of homogeneous layers in DNNs. The bounds are determined by the effective dimensions of the training data embeddings in/out of the layers. This result sheds new light on the architectural requirements for DNNs. Finally, the connection to the multi-task lasso problem suggests a new approach to compressing pre-trained networks.

Abstract: Structured variational autoencoders (SVAEs) combine probabilistic graphical model priors on latent variables, deep neural networks to link latent variables to observed data, and structure-exploiting algorithms for approximate posterior inference. These models are particularly appealing for sequential data, where the prior can capture temporal dependencies. However, despite their conceptual elegance, SVAEs have proven difficult to implement, and more general approaches have been favored in practice. Here, we revisit SVAEs using modern machine learning tools and demonstrate their advantages over more general alternatives in terms of both accuracy and efficiency. First, we develop a modern implementation for hardware acceleration, parallelization, and automatic differentiation of the message passing algorithms at the core of the SVAE. Second, we show that by exploiting structure in the prior, the SVAE learns more accurate models and posterior distributions, which translate into improved performance on prediction tasks. Third, we show how the SVAE can naturally handle missing data, and we leverage this ability to develop a novel, self-supervised training approach. Altogether, these results show that the time is ripe to revisit structured variational autoencoders.