Dimension reduction techniques usually lose information in the sense that reconstructed data are not identical to the original data. However, we argue that it is possible to have reconstructed data identically distributed as the original data, irrespective of the retained dimension or the specific mapping. This can be achieved by learning a distributional model that matches the conditional distribution of data given its low-dimensional latent variables. Motivated by this, we propose Distributional Principal Autoencoder (DPA) that consists of an encoder that maps high-dimensional data to low-dimensional latent variables and a decoder that maps the latent variables back to the data space. For reducing the dimension, the DPA encoder aims to minimise the unexplained variability of the data with an adaptive choice of the latent dimension. For reconstructing data, the DPA decoder aims to match the conditional distribution of all data that are mapped to a certain latent value, thus ensuring that the reconstructed data retains the original data distribution. Our numerical results on climate data, single-cell data, and image benchmarks demonstrate the practical feasibility and success of the approach in reconstructing the original distribution of the data. DPA embeddings are shown to preserve meaningful structures of data such as the seasonal cycle for precipitations and cell types for gene expression.
相關內容
In large-scale regression problems, random Fourier features (RFFs) have significantly enhanced the computational scalability and flexibility of Gaussian processes (GPs) by defining kernels through their spectral density, from which a finite set of Monte Carlo samples can be used to form an approximate low-rank GP. However, the efficacy of RFFs in kernel approximation and Bayesian kernel learning depends on the ability to tractably sample the kernel spectral measure and the quality of the generated samples. We introduce Stein random features (SRF), leveraging Stein variational gradient descent, which can be used to both generate high-quality RFF samples of known spectral densities as well as flexibly and efficiently approximate traditionally non-analytical spectral measure posteriors. SRFs require only the evaluation of log-probability gradients to perform both kernel approximation and Bayesian kernel learning that results in superior performance over traditional approaches. We empirically validate the effectiveness of SRFs by comparing them to baselines on kernel approximation and well-known GP regression problems.
Estimates of causal parameters such as conditional average treatment effects and conditional quantile treatment effects play an important role in real-world decision making. Given this importance, one should ensure these estimators are calibrated. While there is a rich literature on calibrating estimators of non-causal parameters, very few methods have been derived for calibrating estimators of causal parameters, or more generally estimators of quantities involving nuisance parameters. In this work, we provide a general framework for calibrating predictors involving nuisance estimation. We consider a notion of calibration defined with respect to an arbitrary, nuisance-dependent loss $\ell$, under which we say an estimator $\theta$ is calibrated if its predictions cannot be changed on any level set to decrease loss. We prove generic upper bounds on the calibration error of any causal parameter estimate $\theta$ with respect to any loss $\ell$ using a concept called Neyman Orthogonality. Our bounds involve two decoupled terms - one measuring the error in estimating the unknown nuisance parameters, and the other representing the calibration error in a hypothetical world where the learned nuisance estimates were true. We use our bound to analyze the convergence of two sample splitting algorithms for causal calibration. One algorithm, which applies to universally orthogonalizable loss functions, transforms the data into generalized pseudo-outcomes and applies an off-the-shelf calibration procedure. The other algorithm, which applies to conditionally orthogonalizable loss functions, extends the classical uniform mass binning algorithm to include nuisance estimation. Our results are exceedingly general, showing that essentially any existing calibration algorithm can be used in causal settings, with additional loss only arising from errors in nuisance estimation.
In socio-environmental sciences, models are frequently used as tools to represent, understand, project and predict the behaviour of these complex systems. Along the modelling chain, Good Modelling Practices have been evolving that ensure -- amongst others -- that models are transparent and replicable. Whenever such models are represented in software, good modelling meets Good software Practices, such as a tractable development workflow, good code, collaborative development and governance, continuous integration and deployment, and Good Scientific Practices, such as attribution of copyrights and acknowledgement of intellectual property, publication of a software paper and archiving. Too often in existing socio-environmental model software, these practices have been regarded as an add-on to be considered at a later stage only; in fact, many modellers have shied away from publishing their model as open source out of fear that having to add good practices is too demanding. We here argue for making a habit of following a list of simple and not so simple practices early on in the implementation of the model life cycle. We contextualise cherry-picked and hands-on practices for supporting Good Modelling Practices, and we demonstrate their application in the example context of the Viable North Sea fisheries socio-ecological systems model.
A recent trend in Bayesian research has been revisiting generalizations of the likelihood that enable Bayesian inference without requiring the specification of a model for the data generating mechanism. This paper focuses on a Bayesian nonparametric extension of Wedderburn's quasi-likelihood, using Bayesian additive regression trees to model the mean function. Here, the analyst posits only a structural relationship between the mean and variance of the outcome. We show that this approach provides a unified, computationally efficient, framework for extending Bayesian decision tree ensembles to many new settings, including simplex-valued and heavily heteroskedastic data. We also introduce Bayesian strategies for inferring the dispersion parameter of the quasi-likelihood, a task which is complicated by the fact that the quasi-likelihood itself does not contain information about this parameter; despite these challenges, we are able to inject updates for the dispersion parameter into a Markov chain Monte Carlo inference scheme in a way that, in the parametric setting, leads to a Bernstein-von Mises result for the stationary distribution of the resulting Markov chain. We illustrate the utility of our approach on a variety of both synthetic and non-synthetic datasets.
Given a large dataset of many tuples, it is hard for users to pick out their preferred tuples. Thus, the preference query problem, which is to find the most preferred tuples from a dataset, is widely discussed in the database area. In this problem, a utility function is given by the user to evaluate to what extent the user prefers a tuple. However, considering a dataset consisting of N tuples, the existing algorithms need O(N) time to answer a query, or need O(N) time for a cold start to answer a query. The reason is that in a classical computer, a linear time is needed to evaluate the utilities by the utility function for N tuples. In this paper, we discuss the Quantum Preference Query (QPQ) problem, where the dataset is given in a quantum memory, and we use a quantum computer to return the answers. Due to quantum parallelism, the quantum algorithm can theoretically perform better than their classical competitors. We discuss this problem in different kinds of input and output. In the QPQ problem, the input can be a number k or a threshold theta. Given k, the problem is to return k tuples with the highest utilities. Given theta, the problem is to return all the tuples with utilities higher than theta. Also, in QPQ problem, the output can be classical (i.e., a list of tuples) or quantum (i.e., a superposition in quantum bits). We proposed four quantum algorithms to solve the problems in the above four scenarios. We analyze the number of memory accesses needed for each quantum algorithm, which shows that the proposed quantum algorithms are at least quadratically faster than their classical competitors. In our experiments, we show that to answer a QPQ problem, the quantum algorithms achieve up to 1000x improvement in number of memory accesses than their classical competitors, which proved that QPQ problem could be a future direction of the study of preference query problems.
Despite the potential of differentially private data visualization to harmonize data analysis and privacy, research in this area remains relatively underdeveloped. Boxplots are a widely popular visualization used for summarizing a dataset and for comparison of multiple datasets. Consequentially, we introduce a differentially private boxplot. We evaluate its effectiveness for displaying location, scale, skewness and tails of a given empirical distribution. In our theoretical exposition, we show that the location and scale of the boxplot are estimated with optimal sample complexity, and the skewness and tails are estimated consistently. In simulations, we show that this boxplot performs similarly to a non-private boxplot, and it outperforms a boxplot naively constructed from existing differentially private quantile algorithms. Additionally, we conduct a real data analysis of Airbnb listings, which shows that comparable analysis can be achieved through differentially private boxplot visualization.
This book is the result of a seminar in which we reviewed multimodal approaches and attempted to create a solid overview of the field, starting with the current state-of-the-art approaches in the two subfields of Deep Learning individually. Further, modeling frameworks are discussed where one modality is transformed into the other, as well as models in which one modality is utilized to enhance representation learning for the other. To conclude the second part, architectures with a focus on handling both modalities simultaneously are introduced. Finally, we also cover other modalities as well as general-purpose multi-modal models, which are able to handle different tasks on different modalities within one unified architecture. One interesting application (Generative Art) eventually caps off this booklet.
The information bottleneck (IB) method is a technique for extracting information that is relevant for predicting the target random variable from the source random variable, which is typically implemented by optimizing the IB Lagrangian that balances the compression and prediction terms. However, the IB Lagrangian is hard to optimize, and multiple trials for tuning values of Lagrangian multiplier are required. Moreover, we show that the prediction performance strictly decreases as the compression gets stronger during optimizing the IB Lagrangian. In this paper, we implement the IB method from the perspective of supervised disentangling. Specifically, we introduce Disentangled Information Bottleneck (DisenIB) that is consistent on compressing source maximally without target prediction performance loss (maximum compression). Theoretical and experimental results demonstrate that our method is consistent on maximum compression, and performs well in terms of generalization, robustness to adversarial attack, out-of-distribution detection, and supervised disentangling.
Adversarial attack is a technique for deceiving Machine Learning (ML) models, which provides a way to evaluate the adversarial robustness. In practice, attack algorithms are artificially selected and tuned by human experts to break a ML system. However, manual selection of attackers tends to be sub-optimal, leading to a mistakenly assessment of model security. In this paper, a new procedure called Composite Adversarial Attack (CAA) is proposed for automatically searching the best combination of attack algorithms and their hyper-parameters from a candidate pool of \textbf{32 base attackers}. We design a search space where attack policy is represented as an attacking sequence, i.e., the output of the previous attacker is used as the initialization input for successors. Multi-objective NSGA-II genetic algorithm is adopted for finding the strongest attack policy with minimum complexity. The experimental result shows CAA beats 10 top attackers on 11 diverse defenses with less elapsed time (\textbf{6 $\times$ faster than AutoAttack}), and achieves the new state-of-the-art on $l_{\infty}$, $l_{2}$ and unrestricted adversarial attacks.
We introduce an approach for deep reinforcement learning (RL) that improves upon the efficiency, generalization capacity, and interpretability of conventional approaches through structured perception and relational reasoning. It uses self-attention to iteratively reason about the relations between entities in a scene and to guide a model-free policy. Our results show that in a novel navigation and planning task called Box-World, our agent finds interpretable solutions that improve upon baselines in terms of sample complexity, ability to generalize to more complex scenes than experienced during training, and overall performance. In the StarCraft II Learning Environment, our agent achieves state-of-the-art performance on six mini-games -- surpassing human grandmaster performance on four. By considering architectural inductive biases, our work opens new directions for overcoming important, but stubborn, challenges in deep RL.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191