This seminar series aims to provide a platform for young researchers (PhD student or post-doc level) to give invited talks about their research, intending to have a diverse set of talks & speakers on topics related to probabilistic machine learning.

The presentations are virtual talks (~30 min + 15 min Q&A) given live over Zoom. To stimulate discussion and allow for the presentation of ongoing work, the talks will not be recorded unless the speaker separately requests for it. The schedule is in Eastern European Standard Time, click here for a conversion.

Seminar organizers Martin Trapp and Arno Solin

Combining Pseudo-Point and State Space Approximations for Sum-Separable Gaussian Processes

Will Tebbutt (University of Cambridge)

Abstract: State space approximations and pseudo point approximations can be combined in a principled manner to yield scalable approximate inference algorithms for sums of separable Gaussian processes. In this talk, I will: 1. show how this combination can be performed for variational pseudo point approximations via a simple conditional independence result, 2. discuss how existing exact inference algorithms for state space models can be re-purposed for approximate inference, 3. interpret existing related work in light of our work, and 4. briefly discuss some experimental results in a spatio-temporal context. For more info, please see our recent UAI paper.

About the presenter: Will is a PhD student with Rich Turner in the Machine Learning Group at Cambridge, and is interested in probabilistic modelling in general. He is particularly interested in Gaussian processes: how to specify and scale them in large spatio-temporal settings, how best to write software to work with them, and challenges faced in climate science for which they might be helpful.

Automated Augmented Conjugate Inference for Gaussian Processes

Théo Galy-Fajou (TU Berlin)

Abstract: Gaussian Processes are a tool of choice for modelling functions with uncertainties. However, the inference is only tractable analytically for the classical case of regression with Gaussian noise since all other likelihoods are not conjugate with the Gaussian prior. In this talk, I will show how one can transform a large class of likelihoods into conditional conjugate distributions by augmenting them with latent variables. These augmented models have the advantage that, while the posterior inference is still not fully analytic, the full conditionals are! Consequently, one can work easily (and efficiently!) with algorithms like Gibbs sampling or Coordinate Ascent VI (CAVI) and outperform existing inference methods.

About the presenter: Théo Galy-Fajou is a PhD candidate at TU Berlin under the supervision of Prof. Manfred Opper. His work focuses specifically on Gaussian processes and ways to scale them easily to more data and more complex models. He also has a general interest in all approximate Bayesian inference techniques. He is heavily involved in open-source development for inference and visualization techniques of Bayesian methods in the Julia programming language.

Finite Mixture Models Do Not Reliably Learn the Number of Components

Diana Cai (Princeton University)

Abstract: Scientists and engineers are often interested in learning the number of subpopulations (or components) present in a data set. A common suggestion is to use a finite mixture model (FMM) with a prior on the number of components. Past work has shown the resulting FMM component-count posterior is consistent; that is, the posterior concentrates on the true, generating number of components. But consistency requires the assumption that the component likelihoods are perfectly specified, which is unrealistic in practice. In this paper, we add rigour to data-analysis folk wisdom by proving that under even the slightest model misspecification, the FMM component-count posterior diverges: the posterior probability of any particular finite number of components converges to 0 in the limit of infinite data. Contrary to intuition, posterior-density consistency is not sufficient to establish this result. We develop novel sufficient conditions that are more realistic and easily checkable than those common in the asymptotics literature. We illustrate the practical consequences of our theory on simulated and real data.

About the presenter: Diana Cai is a Ph.D. candidate in computer science at Princeton University and is advised by Ryan Adams and Barbara Engelhardt. Her research spans the areas of machine learning and statistics and focuses on developing robust and scalable methods for probabilistic modelling and inference, with an emphasis on flexible, interpretable, and nonparametric machine learning methods. Previously, Diana obtained an A.B. in computer science and statistics from Harvard University, an M.S. in statistics from the University of Chicago, and an M.A. in computer science from Princeton University. Her research is supported in part by a Google PhD Fellowship in Machine Learning.

Causal Decision-making Meets Gaussian Processes

Virginia Aglietti (University of Warwick / DeepMind)

Abstract: Solving decision-making problems in a variety of domains such as healthcare or operations research requires experimentation. By performing interventions, one can understand how a system behaves when an action is taken and thus infer the cause-effect relationships of a phenomenon. Experiments are usually expensive, time-consuming, and may present ethical issues. Therefore, researchers generally have to trade-off cost, time, and other practical considerations to decide which experiments to conduct in order to learn about a system. In this talk, I will present two methodologies that, by linking causal inference, experimental design and Gaussian process (GP) modelling, allow to efficiently learn the causal effects in a graph and identify the optimal intervention to perform. Firstly, I will show how to construct a multi-task causal GP model, the DAG-GP model, which captures the non-trivial correlation structure across different experimental outputs. By sharing experimental information, the DAG-GP model accurately estimates the causal effects in a variety of experimental settings while enabling proper uncertainty quantification. I will then demonstrate how this model, and more generally GP models, can be used within decision-making algorithm to choose experiments to perform. Particularly, I will introduce the Causal Bayesian Optimization algorithm, and I will show how incorporating the knowledge of the causal graph in Bayesian Optimization improves the ability to reason about optimal decision making while decreasing the optimization cost and avoiding suboptimal solutions.

About the presenter: Virginia is a final year PhD student in Statistics at the University of Warwick and a visiting researcher at The Alan Turing Institute. She is supervised by Dr. Theodoros Damoulas. In September, Virginia will join DeepMind as a Research Scientist to work on causal probabilistic models. Virginia is interested in linking probabilistic models, specifically Gaussian processes, and causality to develop algorithms for causal decision making under uncertainty. She has expertise in working with Gaussian processes and variational inference, with a particular focus on models for point processes. In terms of applications, she is particularly interested in spatio-temporal problems in the social sciences.

Equivariant Probabilistic Generative Modelling

Priyank Jaini (University of Amsterdam)

Abstract: In this talk, I will discuss recent work on developing generative models for efficient sampling and inference by incorporating inductive biases in the form of equivariances. I will begin by introducing the Equivariant Stein Variational Gradient Descent algorithm – an equivariant sampling method based on Stein’s identity for sampling from densities with symmetries. Equivariant SVGD explicitly incorporates symmetry information in a density through equivariant kernels, which makes the resultant sampler efficient both in terms of sample complexity and the quality of generated samples. Subsequently, I will demonstrate the use of Equivariant SVGD by defining equivariant energy-based models to model invariant densities that are learned using contrastive divergence. I will then discuss the applications of these equivariant energy models for modelling joint densities in regression and classification tasks for image datasets, many-body particle systems and molecular structure generation. Finally, if time permits, I will touch on methods for sampling using diffusion models and neural transport augmented Monte Carlo methods for more efficient sampling in discrete spaces with applications in denoising, Bayesian posterior sampling, and training light-weight Bayesian quantised neural nets.

About the presenter: Priyank Jaini is a postdoctoral researcher at the University of Amsterdam and Bosch-Delta Lab, working with Prof. Max Welling. Before that, he completed his PhD at the University of Waterloo under the supervision of Prof. Pascal Poupart and Prof. Yaoliang Yu, where he received the doctoral dissertation award from the faculty of Math for his PhD thesis. His research interests lie in building tractable probabilistic models for reasoning under uncertainty. Recently, he has been interested in incorporating inductive biases in the form of symmetries through equivariances in probabilistic modelling and applying them to downstream tasks like molecular generation and modelling many-body particle systems.

Tractable Probabilistic Reasoning for Trustworthy AI

YooJung Choi (UCLA)

Abstract: Automated decision-making systems are increasingly being deployed in areas with personal and societal impacts: from personalized ads to medical diagnosis and criminal justice. This led to growing interest and need for trustworthy AI and ML systems - that is, models that are robust, explainable, fair, and so on. It is important to note that these guarantees only hold with respect to a certain model of the world, with inherent uncertainties. In this talk, I will present how probabilistic modelling and inference, by incorporating a distribution, offer a principled way to handle different kinds of uncertainties when reasoning about decision-making system behaviours. For example, labels in training data may be biased; I will show that probabilistic circuits, a class of tractable probabilistic models (TPMs), can be effective in enforcing and auditing fairness properties by explicitly modelling a latent unbiased label. Another common source of uncertainty is missing values at prediction time, which also leads to fairness and robustness queries that account for this to be computationally hard inference tasks. I will also demonstrate how TPMs can again tractably answer these complex queries.

About the presenter: YooJung Choi is a Ph.D. candidate in the Computer Science Department at UCLA, advised by Prof. Guy Van den Broeck. Her research is broadly in the areas of artificial intelligence and machine learning, with a focus on probabilistic modelling and inference for automated decision-making. In particular, she is interested in developing algorithms for tractable inference of complex queries and characterizing the boundaries of tractable inference. Her work also focuses on applying these results to address fairness, robustness, explainability, and in general, aim towards trustworthy AI/ML.

Probabilistic Inference, Message Passing, and Hybrid Models

Christian Knoll (TU Graz)

Abstract: Probabilistic graphical models are flexible models for representing complex high-dimensional distributions and for incorporating domain knowledge in an intuitive and expressive way. For making probabilistic inference, one often relies on recursive message passing methods. While these methods are efficient for restricted model classes (e.g., for trees), they only serve as approximation methods for more complex models. In this talk, I will show how we can enhance the performance of message passing methods from two opposing angles: i.e., by simplifying the model itself with the utilized inference method in mind and by modifying the inference method with the underlying model in mind. Therefore, I will show how we can advance our understanding of message passing methods by considering them as a dynamic system and by applying tools from system theory. These insights will then suggest various improvements. Recently, we have also complemented message passing methods by neural networks. I will discuss how such hybrid models benefit from the flexibility of neural networks in combination with the implicit underlying model assumptions.

About the presenter: Christian Knoll is a postdoc researcher at the signal processing and speech communications laboratory at the Graz University of Technology. His research interests include machine learning, statistical signal processing, and probabilistic graphical models. He is particularly interested in understanding and improving message passing methods for performing probabilistic inference in graphical models. Recently, he has also been interested in combining graphical models with neural networks.

Score-based Generative Modeling and the Diffusion Schrödinger Bridge

James Thornton (University of Oxford)

Abstract: Progressively applying Gaussian noise transforms complex data distributions to approximately Gaussian. Reversing this dynamic defines a generative model. When the forward noising process is given by a Stochastic Differential Equation (SDE), Song et al. (2021) demonstrate how the time inhomogeneous drift of the associated reverse-time SDE may be estimated using score-matching. A limitation of this approach is that the forward-time SDE must be run for a sufficiently long time for the final distribution to be approximately Gaussian. In contrast, solving the Schrödinger Bridge problem (SB), i.e. an entropy-regularized optimal transport problem on path spaces, yields diffusions which generate samples from the data distribution in finite time. We present Diffusion SB (DSB), an original approximation of the Iterative Proportional Fitting (IPF) procedure to solve the SB problem, and provide theoretical analysis along with generative modeling experiments. The first DSB iteration recovers the methodology proposed by Song et al. (2021), with the flexibility of using shorter time intervals, as subsequent DSB iterations reduce the discrepancy between the final-time marginal of the forward (resp. backward) SDE with respect to the prior (resp. data) distribution. Beyond generative modeling, DSB offers a widely applicable computational optimal transport tool as the continuous state-space analogue of the popular Sinkhorn algorithm (Cuturi, 2013). Joint work with Valentin De Bortoli, Jeremy Heng and Arnaud Doucet.

About the presenter: James is a PhD student at the University of Oxford, supervised by George Deligiannidis and Arnaud Doucet. His research interests are at the intersection of Optimal Transport, sampling methods and machine learning.

Topographic Generative Models Learn Structured Representations

Andy Keller (University of Amsterdam)

Abstract: Topographic generative models can be seen as a class of generative models where the latent variables have an underlying topographic (or spatial) organization which determines their correlation structure. Such structure is widely observed in biological neural networks, however, its computational value is still debated and thus lacks adoption by the deep learning community at large. In this talk, we will describe the statistical motivations behind early topographic generative models like Topographic ICA, and show how such priors can be integrated into modern deep neural networks by introducing the Topographic Variational Autoencoder (TVAE). Further, we will show how topographic representations can be seen as generalized structured representations, and demonstrate how topographic organization over space and time can be leveraged to induce the learning of equivariant sets of features we call capsules. Finally, we will show preliminary results comparing the representations learned by deep TVAEs with FMRI recordings, demonstrating the emergence of localized specialized regions similar to the face area observed in primates.

About the presenter: T. Anderson Keller (Andy) is a fourth-year PhD student supervised by Max Welling at the University of Amsterdam. His work is focused on probabilistic generative models inspired by observations and theories from neuroscience. His current interests broadly include: developing unsupervised methods for structured representation learning (e.g. equivariant & invariant representations), exploring the computational benefits of topographically organized representations, and improving techniques for efficiently training deep latent variable models. His past research includes studying fast-weight recurrent neural networks while part of the Intel AI Lab and developing methods for training unconstrained normalizing flows.

Bayesian Deep Learning with Linearised Neural Networks

Javier Antorán (University of Cambridge)

Abstract: Despite their ubiquitousness in modern data-driven decision-making systems, neural networks are not very well understood. A symptom of this is that network hyperparameters are almost always chosen via cross-validation, an expensive approach that scales poorly in the number of hyperparameters. Additionally, obtaining robust uncertainty estimates for neural network predictions remains an open problem. The probabilistic framework holds the promise of providing both an objective for model selection and reliable uncertainty estimates. However, for the case of neural networks exact probabilistic inference is intractable. This talk introduces the Linearised Laplace approximation for Bayesian deep learning. We examine the assumptions behind linearised Laplace, particularly in conjunction with model selection. We show that these interact poorly with some now-standard features of deep learning—stochastic approximation methods and normalisation layers—and make recommendations for how to better adapt this classic method to the modern setting. We provide Theoretical support of our recommendations and validate them empirically on MLPs, classic CNNs, residual networks with and without normalisation layers, generative autoencoders and transformers. As a case study, we deep dive into Bayesian deep learning methods for tomographic reconstruction. Using linearised Laplace, we construct a probabilistic Deep Image Prior over reconstructed images. Inference in this model allows us to choose U-Net architecture parameters without the need for cross-validation and yields state of the art uncertainty calibration for tomographic reconstruction.

About the presenter: Javier Antorán is a third-year PhD student at the University of Cambridge, supervised by Jose Miguel Hernandez-Lobato and Max Welling. His research focuses on probabilistic modelling and inference. Specifically, Javier’s research spans Bayesian deep learning, Gaussian processes, causal inference and interpretable machine learning. Previously to starting his PhD, Javier worked as a telecommunications engineer developing communications infrastructure for the ATLAS experiment at CERN, and co-founded the startup ARISE, which develops machine learning technology to increase the efficiency of the process in the agricultural sector.

The Coupled Rejection Sampler

Adrien Corenflos (Aalto University)

Abstract: Coupling methods have recently been used to compute unbiased estimates of Markov chains Monte Carlo and particle smoothing expectations. However, in most cases, sampling from couplings has a random run time, the variance of which can be infinite. This behaviour, acceptable in distributed computing, is highly problematic in the parallel computing framework. We propose a limited variance coupled rejection sampling method for sampling from couplings of arbitrary distributions. We show how we can modify the coupled rejection method to propose an ensemble of proposals so as to asymptotically recover a maximal coupling while decreasing the total run time of the algorithm. We then discuss the important special case of coupling Gaussian distributions with different means and covariances, and show how the rejection sampling method can be optimised in this case. We then apply the method to sampling from couplings of Gaussian tails, perform coupled Gibbs sampling, couple parallel resampling algorithms in particle filtering, and couple manifold MALA.

About the presenter: Adrien Corenflos is a PhD student at Aalto university under Simo Särkkä. Prior to this, he worked as a quantitative researcher at JP Morgan, UK. His research interests typically orbit around Monte Carlo methods for machine learning, with an emphasis on sequential Monte Carlo.

Low-Cost Bayesian Methods for Fixing Neural Nets' Overconfidence

Agustinus Kristiadi (University of Tübingen)

Abstract: Well-calibrated predictive uncertainty of neural networks—essentially making them know when they don’t know—is paramount in safety-critical applications. However, deep neural networks are overconfident in the region both far away and near the training data. In our works, we study Bayesian neural networks (BNNs) and their extensions to mitigate this issue. First, we show that being Bayesian, even just at the last layer and in a post-hoc manner, helps mitigate overconfidence in deep ReLU classifiers. Then, we provide a cost-effective Gaussian-process extension to ReLU BNNs that provides a guarantee that ReLU nets will never be overconfident in the region far from the data. Furthermore, we propose two ways of improving the calibration of general BNNs in the out-of-distribution (OOD) regions near the data by (i) training the uncertainty of Laplace approximations and (ii) by leveraging OOD data during training. Finally, we provide a simple library, laplace-torch, to facilitate modern arts of Laplace approximations in deep learning. This library gives users a way to turn a standard pre-trained deep net into a BNN in a cost-efficient manner.

About the presenter: Agustinus Kristiadi is a last-year Ph.D. student at the University of Tübingen, under the supervision of Philipp Hennig. Before this, he studied computer science at the University of Bonn under Asja Fischer. His current interest is in the intersection between Bayesian deep learning and Riemannian geometry.

Embedded-model flows: Combining the inductive biases of model-free deep learning and explicit probabilistic modeling

Gianluigi Silvestri (OnePlanet Research Center and Donders Institute for Brain, Cognition and Behaviour)

Abstract: Normalizing flows have shown great success as general-purpose density estimators. However, many real-world applications require the use of domain-specific knowledge, which normalizing flows cannot readily incorporate. We propose embedded-model flows (EMF), which alternate general-purpose transformations with structured layers that embed domain-specific inductive biases. These layers are automatically constructed by converting user-specified differentiable probabilistic models into equivalent bijective transformations. We also introduce gated structured layers, which allow bypassing the parts of the models that fail to capture the statistics of the data. We demonstrate that EMFs can be used to induce desirable properties such as multimodality, hierarchical coupling and continuity. Furthermore, we show that EMFs enable a high-performance form of variational inference where the structure of the prior model is embedded in the variational architecture. In our experiments, we show that this approach outperforms state-of-the-art methods in common structured inference problems.

About the presenter: Gianluigi Silvestri is a PhD candidate in Probabilistic Machine Learning at OnePlanet Research Center and Donders Institute for Brain, Cognition and Behaviour, in Nijmegen, the Netherlands. His research interests include Variational Inference, Normalizing Flows and Bayesian Reinforcement Learning.

Parameter elimination in particle Gibbs sampling

Anna Wigren (Uppsala University)

Abstract: Bayesian joint parameter and state inference in non-linear state-space models is a difficult problem due to the often high-dimensional state sequence. Particle Gibbs (PG) is well-suited for solving this type of inference problem but produces correlated samples. In this talk, I describe how the correlation can be reduced by marginalizing out one or more parameters from the state update when conjugacy relations exist between the parameter prior and the complete data likelihood. Deriving the marginalized conjugacy relations is often time-consuming, but probabilistic programming can be employed to automate the process. I also introduce a marginalized PG sampler for multiple time series described by a common state-space model structure, where subsets of the parameters are shared between different models. The spread of mosquito-borne diseases, where some parameters are location-specific, and some are disease-specific, is one example. In theory, it is possible to update all models concurrently, but sequential Monte Carlo becomes inefficient as the number of time series increases. Our suggested marginalized PG sampler instead updates one model at a time, conditioned on the remaining datasets, and can be formulated in a modular fashion that greatly facilitates its implementation.

About the presenter: Anna Wigren is a final year PhD student at the Division of Systems and Control at Uppsala University, supervised by Fredrik Lindsten (Linköping University). Her main research interests include sequential Monte Carlo and Markov chain Monte Carlo methods.

Adaptive Design in Real Time

Desi Ivanova (University of Oxford)

Abstract: Designing sequences of adaptive experiments to maximise the information gathered about an underlying process is a key challenge in science and engineering. Bayesian Experimental Design (BED) is a powerful mathematical framework for tackling the optimal design problem. Despite the huge potential of obtaining information more quickly and efficiently, the widespread adoption of adaptive BED has been severely limited by the costly computations required at each experiment iteration. In this talk, I’ll present a new method, called Deep Adaptive Design (DAD), that alleviates this problem. DAD marks a critical change from previous BED methods in that it optimises a policy instead of individual designs during the experiment. The policy is parametrised by a neural network, taking as inputs past data and returning the design to use at the next experiment iteration. Using a single pass through the network, DAD enables quick and adaptive design decisions in real time.

About the presenter: Desi Ivanova is a second-year grad student at the University of Oxford, working with Tom Rainforth and Yee Whye Teh. Her research interests include probabilistic machine learning and inference with applications to experimental design, causality and data compression. Before joining Oxford, Desi worked as a quantitative researcher at Goldman Sachs.

Aspects of modelling CoVID-19: Understanding and quantifying the uncertainty

Swapnil Mishra (University of Copenhagen)

Abstract: Despite trends in modern medicine and epidemiological control, the risk for novel outbreaks and previously existing pathogens is currently greater than ever. Indeed, the current outbreak of SARS-CoV-2 has exposed the need for precise, robust, and principled mathematical modelling of disease outbreaks that can perform well with noisy and potentially biased data. To tackle these challenges, I will present a unifying view of modelling infectious diseases that contributes to the new understanding of the spread of the diseases and their epidemiological properties. The unified framework allows flexible probabilistic models that are capable of fitting complex and noisy data from different sources. I will touch upon how the new unified framework, built using Stan (numpyro), has helped us to characterize the initial spread of SARS-CoV-2 and quantify the altered epidemiological characteristics of various ‘variants of concerns’ (VOCs).

About the presenter: I am an Assistant Professor at the University of Copenhagen (UCPH), where I am primarily working at the intersection of public health, machine learning and Bayesian modelling. Before this, I was fortunate enough to spend my post-doc years at the School of Public Health, Imperial College London, where I worked primarily with Professor Samir Bhatt and Dr Seth Flaxman. I am working with colleagues at Imperial College London, the University of Oxford and UCPH to model the spread of COVID-19. I did my Ph.D. at the Research School of Computer Science, The Australian National University, under the supervision of Professor Lexing Xie and Dr Marian-Andrei Rizoiu. I have also worked with Professor Wray Buntine at Monash University.

Transformed Gaussian Processes to specify non-stationary function priors

Juan Maroñas (Universidad Autónoma of Madrid)

Abstract: In this talk, I will introduce the Transformed Gaussian Processes, a stochastic process specified by transforming samples from a Gaussian process using an invertible transformation (warping function). These processes can be easily made non-stationary by parameterizing the warping function through an input-dependent transformation. I show how this is achieved with a Bayesian Neural Network implemented with Monte Carlo dropout with the additional benefit of incorporating uncertainties, effectively regularizing the model. This new model can match the performance of a Deep Gaussian Process at a fraction of its cost and also allow us to incorporate inductive biases in the function that we are trying to model (e.g. positive constraints), among other benefits. Training and predictions can be scaled using a sparse variational inference algorithm. We also show how the basic idea of Transformed Gaussian Processes can be used to create a set of C dependent function priors which can provide similar or better results than an SVGP in classification problems with a big number of classes but one order of magnitude faster.

About the presenter: Since February 2022, I am a Post-doctoral researcher at Universidad Autónoma of Madrid, working with Daniel Hernández Lobato. I defended my PhD in January 2022 from Universidad Politécnica de Valencia. My research interests are Bayesian modelling (Gaussian Processes, Bayesian Neural Networks and hierarchical latent variable models) and the different ways of performing inference. I am also interested in deep generative models, Bayes decision theory and model calibration. Previous to this, I have trained Deep Convolutional Neural Networks for computer vision in different domains and also worked on other application problems such as speech enhancement using deep learning. In the next years, I am planning to study probabilistic circuits and stochastic differential equations in depth.

A Kernel Stein Test for Comparing Latent Variable Models

Heishiro Kanagawa (Gatsby Computational Neuroscience Unit, University College London)

Abstract: I will describe a kernel-based nonparametric test of relative goodness of fit, where the goal is to compare two models, both of which may have unobserved latent variables, such that the marginal distributions of the observed variables are intractable. Given the premise that “all models are wrong,” the goal of the test is to determine whether one model significantly outperforms the other in respect of a reference data sample. The test generalises earlier kernel Stein discrepancy (KSD) tests to the case of latent variable models, a much more general class than the fully observed models treated previously. The new test, with a properly calibrated threshold, has a well-controlled type-I error. In the case of models with low-dimensional latent structure and high-dimensional observations, our test significantly outperforms the relative maximum mean discrepancy test, which is based on samples from the models, and does not exploit the latent structure. I will illustrate the test on probabilistic topic models of arXiv articles.

About the presenter: Heishiro Kanagawa is a PhD student at Gatsby Computational Neuroscience Unit supervised by Arthur Gretton, and joining Newcastle University to work with Chris Oates. He is interested in evaluating machine learning and statistical models and developing diagnostic tools.

Bayesian Optimization over Discrete and Mixed Spaces via Probabilistic Reparameterization

Sam Daulton (Meta / University of Oxford)

Abstract: Optimizing expensive-to-evaluate black-box functions of discrete (and potentially continuous) design parameters is a ubiquitous problem in science and engineering applications. Bayesian optimization (BO) is a popular sample-efficient method that selects promising designs to evaluate by optimizing an acquisition function (AF) over some domain based on a probabilistic surrogate model. However, maximizing the AF over mixed or high-cardinality discrete search spaces is challenging as we cannot use standard gradient-based methods or evaluate the AF at every point in the search space. To address this issue, we propose using probabilistic reparameterization (PR). Instead of directly optimizing the AF over the search space containing discrete parameters, we instead maximize the expectation of the AF over a probability distribution defined by continuous parameters. We prove that under suitable reparameterizations, the BO policy that maximizes the probabilistic objective is the same as that which maximizes the AF, and therefore, PR enjoys the same regret bounds as the underlying AF. Moreover, our approach admits provably converges to a stationary point of the probabilistic objective under gradient ascent using scalable, unbiased estimators of both the probabilistic objective and its gradient, and therefore, as the number of starting points and gradient steps increases, our approach will recover of a maximizer of the AF (an often-neglected requisite for commonly used BO regret bounds). We validate our approach empirically and demonstrate state-of-the-art optimization performance on many real-world applications. PR is complementary to (and benefits) recent work and naturally generalizes to settings with multiple objectives and black-box constraints.

About the presenter: Sam Daulton is a research scientist at Meta, a DPhil candidate in machine learning at the University of Oxford, and co-creator of BoTorch—an open-source library for Bayesian optimization research. Sam works with Eytan Bakshy and Max Balandat at Meta and Mike Osborne at Oxford. His current research focuses on methods for Bayesian optimization in challenging scenarios. Previously, Sam worked with Finale Doshi-Velez at Harvard University on efficient and robust methods for transfer learning.

Mitigating Modality Collapse in Multimodal VAEs via Impartial Optimization

Adrián Javaloy (Saarland University)

Abstract: A number of variational autoencoders (VAEs) have recently emerged with the aim of modelling multimodal data, e.g., to jointly model images and their corresponding captions. Still, multimodal VAEs tend to focus solely on a subset of the modalities, e.g., by fitting the image while neglecting the caption. We refer to this limitation as modality collapse. In this presentation, I argue that this effect is a consequence of modality-specific gradients conflicting during the training of multimodal VAEs. After this talk, you will be able to detect which parts of your model’s computational graph can suffer from gradients conflict (which I call impartiality blocks), as well as how to leverage existing gradient-conflict solutions from multitask learning to mitigate modality collapse. In order words, you will learn how to encourage impartial optimization across modalities. The framework I introduce is general, and we have successfully applied it to several multimodal VAE models, losses, and datasets from the literature, and empirically showed that it significantly improves the reconstruction performance, conditional generation, and coherence of the latent space across modalities.

About the presenter: I am a last-year PhD student at the University of Saarland under the supervision of Isabel Valera, where I primarily work on probabilistic machine learning models and their training dynamics (and challenges), specially when the training data comprises multiple modalities (such as images and their captions, or tabular data, where different columns represent fundamentally different quantities). My interests are broad however, and I try to stay tuned with the machine learning literature without losing my own sanity. Before moving to Saarland, I started my PhD at the Max-Planck Institute for Intelligent Systems with Isabel Valera, and previously I pursued a double-bachelor on Computer Science Engineering and Mathematics at the University of Murcia.

Learning to Dynamically Optimise Algorithms

André Biedenkapp (Freiburg University)

Abstract: TBA

About the presenter:


Antonio Horta Ribeiro (Uppsala University)

Abstract: TBA

About the presenter:


Emile Mathieu (University of Cambridge)

Abstract: TBA

About the presenter:


Vincent Fortuin (University of Cambridge)

Abstract: TBA

About the presenter:


Beliz Gunel (Stanford University / Google AI)

Abstract: TBA

About the presenter:


Harita Dellaporta (University of Warwick)

Abstract: TBA

About the presenter:


Benjie Wang (University of Oxford)

Abstract: TBA

About the presenter:


Peter Nicholas Krämer (University of Tübingen)

Abstract: TBA

About the presenter: