We propose a probabilistic framework for domain adaptation that blends both generative and discriminative modeling in a principled way. Under this framework, generative and discriminative models correspond to specific choices of the prior over parameters. This provides us a very general way to interpolate between generative and discriminative extremes through different choices of priors. By maximizing both the marginal and the conditional log-likelihoods, models derived from this framework can use both labeled instances from the source domain as well as unlabeled instances from both source and target domains. Under this framework, we show that the popular reconstruction loss of autoencoder corresponds to an upper bound of the negative marginal log-likelihoods of unlabeled instances, where marginal distributions are given by proper kernel density estimations. This provides a way to interpret the empirical success of autoencoders in domain adaptation and semi-supervised learning. We instantiate our framework using neural networks, and build a concrete model, DAuto. Empirically, we demonstrate the effectiveness of DAuto on text, image and speech datasets, showing that it outperforms related competitors when domain adaptation is possible.

I chose the Bay Area housing price dataset that was sourced from Bay Area Home Sales Database and Zillow. This dataset was based on the homes sold between January 2013 and December 2015. It has many characteristics of learning. The dataset can be downloaded from here. There are several features that we do not need, such as "info", "z_address", "zipcode"(We have "neighborhood" as a location variable), "zipid" and "zestimate"(This is the price estimated by Zillow, we don't want our model to be affected by this).

Research fields usually follow the practice of categorizing continuous predictor variables, and they are the same who mostly use ANOVA. They often do it through median splits, the high value above the median and the low values below the median. The way out of this dilemma is to be able to conclude whether to treat an independent variable as categorical or continuous. Data analysts are empowered to find real results which otherwise they might miss, is by knowing when it is appropriate, followed with the understanding of how it will affect the interpretation of parameters. Let's understand and accept the fact that general linear model is not concerned if the predictor you used is continuous or categorical. But you as a data analyst should choose the information you need from the analysis based on the coding of the predictor.

Social media provide a low-cost alternative source for public health surveillance and health-related classification plays an important role to identify useful information. We summarized the recent classification methods using social media in public health. These methods rely on bag-of-words (BOW) model and have difficulty grasping the semantic meaning of texts. Unlike these methods, we present a word embedding based clustering method. Word embedding is one of the strongest trends in Natural Language Processing (NLP) at this moment. It learns the optimal vectors from surrounding words and the vectors can represent the semantic information of words.

Conjugate gradient methods are a class of important methods for solving linear equations and nonlinear optimization. In our work, we propose a new stochastic conjugate gradient algorithm with variance reduction (CGVR) and prove its linear convergence with the Fletcher and Revves method for strongly convex and smooth functions. We experimentally demonstrate that the CGVR algorithm converges faster than its counterparts for six large-scale optimization problems that may be convex, non-convex or non-smooth, and its AUC (Area Under Curve) performance with $L2$-regularized $L2$-loss is comparable to that of LIBLINEAR but with significant improvement in computational efficiency.

Energy statistics was proposed by Sz\'{e}kely in the 80's inspired by the Newtonian gravitational potential from classical mechanics, and it provides a hypothesis test for equality of distributions. It was further generalized from Euclidean spaces to metric spaces of strong negative type, and more recently, a connection with reproducing kernel Hilbert spaces (RKHS) was established. Here we consider the clustering problem from an energy statistics theory perspective, providing a precise mathematical formulation yielding a quadratically constrained quadratic program (QCQP) in the associated RKHS, thus establishing the connection with kernel methods. We show that this QCQP is equivalent to kernel $k$-means optimization problem once the kernel is fixed. These results imply a first principles derivation of kernel $k$-means from energy statistics. However, energy statistics fixes a family of standard kernels. Furthermore, we also consider a weighted version of energy statistics, making connection to graph partitioning problems. To find local optimizers of such QCQP we propose an iterative algorithm based on Hartigan's method, which in this case has the same computational cost as kernel $k$-means algorithm, based on Lloyd's heuristic, but usually with better clustering quality. We provide carefully designed numerical experiments showing the superiority of the proposed method compared to kernel $k$-means, spectral clustering, standard $k$-means, and Gaussian mixture models in a variety of settings.

Modern large scale machine learning applications require scaling stochastic optimization algorithms to distributed computational architectures. A key bottleneck is the communication overhead for exchanging information among different workers. For example, we have n training data distributed on M workers, and each of them owns its local copy of the model parameter vector. In the synchronized stochastic gradient method, each worker processes a random minibatch of its training data, and then the local updates are synchronized by making an All-Reduce step, which aggregates stochastic gradients from all workers, and taking a Broadcast step that transmits the updated parameter vector back to all workers. The process is repeated until an appropriate convergence criterion is met. An important factor that may significantly slow down any optimization algorithm is the communication cost among workers.

This work addresses the problem of segmentation in time series data with respect to a statistical parameter of interest in Bayesian models. It is common to assume that the parameters are distinct within each segment. As such, many Bayesian change point detection models do not exploit the segment parameter patterns, which can improve performance. This work proposes a Bayesian mean-shift change point detection algorithm that makes use of repetition in segment parameters, by introducing segment class labels that utilise a Dirichlet process prior. The performance of the proposed approach was assessed on both synthetic and real world data, highlighting the enhanced performance when using parameter labelling.

Graph embedding methods represent nodes in a continuous vector space, preserving information from the graph (e.g. by sampling random walks). There are many hyper-parameters to these methods (such as random walk length) which have to be manually tuned for every graph. In this paper, we replace random walk hyper-parameters with trainable parameters that we automatically learn via backpropagation. In particular, we learn a novel attention model on the power series of the transition matrix, which guides the random walk to optimize an upstream objective. Unlike previous approaches to attention models, the method that we propose utilizes attention parameters exclusively on the data (e.g. on the random walk), and not used by the model for inference. We experiment on link prediction tasks, as we aim to produce embeddings that best-preserve the graph structure, generalizing to unseen information. We improve state-of-the-art on a comprehensive suite of real world datasets including social, collaboration, and biological networks. Adding attention to random walks can reduce the error by 20% to 45% on datasets we attempted. Further, our learned attention parameters are different for every graph, and our automatically-found values agree with the optimal choice of hyper-parameter if we manually tune existing methods.

We present an algorithm for classification tasks on big data. Experiments conducted as part of this study indicate that the algorithm can be as accurate as ensemble methods such as random forests or gradient boosted trees. Unlike ensemble methods, the models produced by the algorithm can be easily interpreted. The algorithm is based on a divide and conquer strategy and consists of two steps. The first step consists of using a decision tree to segment the large dataset. By construction, decision trees attempt to create homogeneous class distributions in their leaf nodes. However, non-homogeneous leaf nodes are usually produced. The second step of the algorithm consists of using a suitable classifier to determine the class labels for the non-homogeneous leaf nodes. The decision tree segment provides a coarse segment profile while the leaf level classifier can provide information about the attributes that affect the label within a segment.