We aim for source-free domain adaptation, where the task is to deploy a model pre-trained on source domains to target domains. The challenges stem from the distribution shift from the source to the target domain, coupled with the unavailability of any source data and labeled target data for optimization. Rather than fine-tuning the model by updating the parameters, we propose to perturb the source model to achieve adaptation to target domains. We introduce perturbations into the model parameters by variational Bayesian inference in a probabilistic framework. By doing so, we can effectively adapt the model to the target domain while largely preserving the discriminative ability. Importantly, we demonstrate the theoretical connection to learning Bayesian neural networks, which proves the generalizability of the perturbed model to target domains. To enable more efficient optimization, we further employ a parameter sharing strategy, which substantially reduces the learnable parameters compared to a fully Bayesian neural network. Our model perturbation provides a new probabilistic way for domain adaptation which enables efficient adaptation to target domains while maximally preserving knowledge in source models. Experiments on several source-free benchmarks under three different evaluation settings verify the effectiveness of the proposed variational model perturbation for source-free domain adaptation.

The Bayesian transformed Gaussian process (BTG) model, proposed by Kedem and Oliviera, is a fully Bayesian counterpart to the warped Gaussian process (WGP) and marginalizes out a joint prior over input warping and kernel hyperparameters. This fully Bayesian treatment of hyperparameters often provides more accurate regression estimates and superior uncertainty propagation, but is prohibitively expensive. The BTG posterior predictive distribution, itself estimated through high-dimensional integration, must be inverted in order to perform model prediction. To make the Bayesian approach practical and comparable in speed to maximum-likelihood estimation (MLE), we propose principled and fast techniques for computing with BTG. Our framework uses doubly sparse quadrature rules, tight quantile bounds, and rank-one matrix algebra to enable both fast model prediction and model selection. These scalable methods allow us to regress over higher-dimensional datasets and apply BTG with layered transformations that greatly improve its expressibility. We demonstrate that BTG achieves superior empirical performance over MLE-based models.

Volatility clustering is a common phenomenon in financial time series. Typically, linear models can be used to describe the temporal autocorrelation of the (logarithmic) variance of returns. Considering the difficulty in estimating this model, we construct a Dynamic Bayesian Network, which utilizes the conjugate prior relation of normal-gamma and gamma-gamma, so that its posterior form locally remains unchanged at each node. This makes it possible to find approximate solutions using variational methods quickly. Furthermore, we ensure that the volatility expressed by the model is an independent incremental process after inserting dummy gamma nodes between adjacent time steps. We have found that this model has two advantages: 1) It can be proved that it can express heavier tails than Gaussians, i.e., have positive excess kurtosis, compared to popular linear models. 2) If the variational inference(VI) is used for state estimation, it runs much faster than Monte Carlo(MC) methods since the calculation of the posterior uses only basic arithmetic operations. And its convergence process is deterministic. We tested the model, named Gam-Chain, using recent Crypto, Nasdaq, and Forex records of varying resolutions. The results show that: 1) In the same case of using MC, this model can achieve comparable state estimation results with the regular lognormal chain. 2) In the case of only using VI, this model can obtain accuracy that are slightly worse than MC, but still acceptable in practice; 3) Only using VI, the running time of Gam-Chain, in general case, can be reduced to below 5% of that based on the lognormal chain via MC.

The page presentation biases in the information retrieval system, especially on the click behavior, is a well-known challenge that hinders improving ranking models' performance with implicit user feedback. Unbiased Learning to Rank~(ULTR) algorithms are then proposed to learn an unbiased ranking model with biased click data. However, most existing algorithms are specifically designed to mitigate position-related bias, e.g., trust bias, without considering biases induced by other features in search result page presentation(SERP). For example, the multimedia type may generate attractive bias. Unfortunately, those biases widely exist in industrial systems and may lead to an unsatisfactory search experience. Therefore, we introduce a new problem, i.e., whole-page Unbiased Learning to Rank(WP-ULTR), aiming to handle biases induced by whole-page SERP features simultaneously. It presents tremendous challenges. For example, a suitable user behavior model (user behavior hypothesis) can be hard to find; and complex biases cannot be handled by existing algorithms. To address the above challenges, we propose a Bias Agnostic whole-page unbiased Learning to rank algorithm, BAL, to automatically discover and mitigate the biases from multiple SERP features with no specific design. Experimental results on a real-world dataset verify the effectiveness of the BAL.

We revisit the challenging problem of training Gaussian-Bernoulli restricted Boltzmann machines (GRBMs), introducing two innovations. We propose a novel Gibbs-Langevin sampling algorithm that outperforms existing methods like Gibbs sampling. We propose a modified contrastive divergence (CD) algorithm so that one can generate images with GRBMs starting from noise. This enables direct comparison of GRBMs with deep generative models, improving evaluation protocols in the RBM literature. Moreover, we show that modified CD and gradient clipping are enough to robustly train GRBMs with large learning rates, thus removing the necessity of various tricks in the literature. Experiments on Gaussian Mixtures, MNIST, FashionMNIST, and CelebA show GRBMs can generate good samples, despite their single-hidden-layer architecture. Our code is released at: \url{https://github.com/lrjconan/GRBM}.

We explore using neural operators, or neural network representations of nonlinear maps between function spaces, to accelerate infinite-dimensional Bayesian inverse problems (BIPs) with models governed by nonlinear parametric partial differential equations (PDEs). Neural operators have gained significant attention in recent years for their ability to approximate the parameter-to-solution maps defined by PDEs using as training data solutions of PDEs at a limited number of parameter samples. The computational cost of BIPs can be drastically reduced if the large number of PDE solves required for posterior characterization are replaced with evaluations of trained neural operators. However, reducing error in the resulting BIP solutions via reducing the approximation error of the neural operators in training can be challenging and unreliable. We provide an a priori error bound result that implies certain BIPs can be ill-conditioned to the approximation error of neural operators, thus leading to inaccessible accuracy requirements in training. To reliably deploy neural operators in BIPs, we consider a strategy for enhancing the performance of neural operators, which is to correct the prediction of a trained neural operator by solving a linear variational problem based on the PDE residual. We show that a trained neural operator with error correction can achieve a quadratic reduction of its approximation error, all while retaining substantial computational speedups of posterior sampling when models are governed by highly nonlinear PDEs. The strategy is applied to two numerical examples of BIPs based on a nonlinear reaction--diffusion problem and deformation of hyperelastic materials. We demonstrate that posterior representations of the two BIPs produced using trained neural operators are greatly and consistently enhanced by error correction.

In computational histopathology algorithms now outperform humans on a range of tasks, but to date none are employed for automated diagnoses in the clinic. Before algorithms can be involved in such high-stakes decisions they need to "know when they don't know", i.e., they need to estimate their predictive uncertainty. This allows them to defer potentially erroneous predictions to a human pathologist, thus increasing their safety. Here, we evaluate the predictive performance and calibration of several uncertainty estimation methods on clinical histopathology data. We show that a distance-aware uncertainty estimation method outperforms commonly used approaches, such as Monte Carlo dropout and deep ensembles. However, we observe a drop in predictive performance and calibration on novel samples across all uncertainty estimation methods tested. We also investigate the use of uncertainty thresholding to reject out-of-distribution samples for selective prediction. We demonstrate the limitations of this approach and suggest areas for future research.

In sponsored search advertising (SSA), advertisers need to select keywords and determine matching types for selected keywords simultaneously, i.e., keyword targeting. An optimal keyword targeting strategy guarantees reaching the right population effectively. This paper aims to address the keyword targeting problem, which is a challenging task because of the incomplete information of historical advertising performance indices and the high uncertainty in SSA environments. First, we construct a data distribution estimation model and apply a Markov Chain Monte Carlo method to make inference about unobserved indices (i.e., impression and click-through rate) over three keyword matching types (i.e., broad, phrase and exact). Second, we formulate a stochastic keyword targeting model (BB-KSM) combining operations of keyword selection and keyword matching to maximize the expected profit under the chance constraint of the budget, and develop a branch-and-bound algorithm incorporating a stochastic simulation process for our keyword targeting model. Finally, based on a realworld dataset collected from field reports and logs of past SSA campaigns, computational experiments are conducted to evaluate the performance of our keyword targeting strategy. Experimental results show that, (a) BB-KSM outperforms seven baselines in terms of profit; (b) BB-KSM shows its superiority as the budget increases, especially in situations with more keywords and keyword combinations; (c) the proposed data distribution estimation approach can effectively address the problem of incomplete performance indices over the three matching types and in turn significantly promotes the performance of keyword targeting decisions. This research makes important contributions to the SSA literature and the results offer critical insights into keyword management for SSA advertisers.

Multimodal distributions of some physics based model parameters are often encountered in engineering due to different situations such as a change in some environmental conditions, and the presence of some types of damage and nonlinearity. In statistical model updating, for locally identifiable parameters, it can be anticipated that multi-modal posterior distributions would be found. The full characterization of these multi-modal distributions is important as methodologies for structural condition monitoring in structures are frequently based in the comparison of the damaged and healthy models of the structure. The characterization of posterior multi-modal distributions using state-of-the-art sampling techniques would require a large number of simulations of expensive to run physics-based models. Therefore, when a limited number of simulations can be run, as it often occurs in engineering, the traditional sampling techniques would not be able to capture accurately the multimodal distributions. This could potentially lead to large numerical errors when assessing the performance of an engineering structure under uncertainty.

Machine learning (ML) formalizes the problem of getting computers to learn from experience as optimization of performance according to some metric(s) on a set of data examples. This is in contrast to requiring behaviour specified in advance (e.g. by hard-coded rules). Formalization of this problem has enabled great progress in many applications with large real-world impact, including translation, speech recognition, self-driving cars, and drug discovery. But practical instantiations of this formalism make many assumptions - for example, that data are i.i.d.: independent and identically distributed - whose soundness is seldom investigated. And in making great progress in such a short time, the field has developed many norms and ad-hoc standards, focused on a relatively small range of problem settings. As applications of ML, particularly in artificial intelligence (AI) systems, become more pervasive in the real world, we need to critically examine these assumptions, norms, and problem settings, as well as the methods that have become de-facto standards. There is much we still do not understand about how and why deep networks trained with stochastic gradient descent are able to generalize as well as they do, why they fail when they do, and how they will perform on out-of-distribution data. In this thesis I cover some of my work towards better understanding deep net generalization, identify several ways assumptions and problem settings fail to generalize to the real world, and propose ways to address those failures in practice.