We study contextual stochastic optimization problems, where we leverage rich auxiliary observations (e.g., product characteristics) to improve decision making with uncertain variables (e.g., demand). We show how to train forest decision policies for this problem by growing trees that choose splits to directly optimize the downstream decision quality, rather than splitting to improve prediction accuracy as in the standard random forest algorithm. We realize this seemingly computationally intractable problem by developing approximate splitting criteria that utilize optimization perturbation analysis to eschew burdensome re-optimization for every candidate split, so that our method scales to large-scale problems. We prove that our splitting criteria consistently approximate the true risk and that our method achieves asymptotic optimality. We extensively validate our method empirically, demonstrating the value of optimization-aware construction of forests and the success of our efficient approximations. We show that our approximate splitting criteria can reduce running time hundredfold, while achieving performance close to forest algorithms that exactly re-optimize for every candidate split.

While theoretically appealing, the application of the Wasserstein distance to large-scale machine learning problems has been hampered by its prohibitive computational cost. The sliced Wasserstein distance and its variants improve the computational efficiency through random projection, yet they suffer from low projection efficiency because the majority of projections result in trivially small values. In this work, we propose a new family of distance metrics, called augmented sliced Wasserstein distances (ASWDs), constructed by first mapping samples to higher-dimensional hypersurfaces parameterized by neural networks. It is derived from a key observation that (random) linear projections of samples residing on these hypersurfaces would translate to much more flexible projections in the original sample space, so they can capture complex structures of the data distribution. We show that the hypersurfaces can be optimized by gradient ascent efficiently. We provide the condition under which the ASWD is a valid metric and show that this can be obtained by an injective neural network architecture. Numerical results demonstrate that the ASWD significantly outperforms other Wasserstein variants for both synthetic and real-world problems.

Markov chain Monte Carlo (MCMC) algorithms offer various strategies for sampling; the Hamiltonian Monte Carlo (HMC) family of samplers are MCMC algorithms which often exhibit improved mixing properties. The recently introduced magnetic HMC, a generalization of HMC motivated by the physics of particles influenced by magnetic field forces, has been demonstrated to improve the performance of HMC. In many applications, one wishes to sample from a distribution restricted to a constrained set, often manifested as an embedded manifold (for example, the surface of a sphere). We introduce magnetic manifold HMC, an HMC algorithm on embedded manifolds motivated by the physics of particles constrained to a manifold and moving under magnetic field forces. We discuss the theoretical properties of magnetic Hamiltonian dynamics on manifolds, and introduce a reversible and symplectic integrator for the HMC updates. We demonstrate that magnetic manifold HMC produces favorable sampling behaviors relative to the canonical variant of manifold-constrained HMC.

In this paper, we analyze and extend an online learning framework known as Context-Attentive Bandit, motivated by various practical applications, from medical diagnosis to dialog systems, where due to observation costs only a small subset of a potentially large number of context variables can be observed at each iteration; however, the agent has a freedom to choose which variables to observe. We derive a novel algorithm, called Context-Attentive Thompson Sampling (CATS), which builds upon the Linear Thompson Sampling approach, adapting it to Context-Attentive Bandit setting. We provide a theoretical regret analysis and an extensive empirical evaluation demonstrating advantages of the proposed approach over several baseline methods on a variety of real-life datasets.

One major task of spoken language understanding (SLU) in modern personal assistants is to extract semantic concepts from an utterance, called slot filling. Although existing slot filling models attempted to improve extracting new concepts that are not seen in training data, the performance in practice is still not satisfied. Recent research collected question and answer annotated data to learn what is unknown and should be asked, yet not practically scalable due to the heavy data collection effort. In this paper, we incorporate softmax-based slot filling neural architectures to model the sequence uncertainty without question supervision. We design a Dirichlet Prior RNN to model high-order uncertainty by degenerating as softmax layer for RNN model training. To further enhance the uncertainty modeling robustness, we propose a novel multi-task training to calibrate the Dirichlet concentration parameters. We collect unseen concepts to create two test datasets from SLU benchmark datasets Snips and ATIS. On these two and another existing Concept Learning benchmark datasets, we show that our approach significantly outperforms state-of-the-art approaches by up to 8.18%. Our method is generic and can be applied to any RNN or Transformer based slot filling models with a softmax layer.

The Smart grid (SG), generally known as the next-generation power grid emerged as a replacement for ill-suited power systems in the 21st century. It is in-tegrated with advanced communication and computing capabilities, thus it is ex-pected to enhance the reliability and the efficiency of energy distribution with minimum effects. With the massive infrastructure it holds and the underlying communication network in the system, it introduced a large volume of data that demands various techniques for proper analysis and decision making. Big data analytics, machine learning (ML), and deep learning (DL) plays a key role when it comes to the analysis of this massive amount of data and generation of valuable insights. This paper explores and surveys the Smart grid architectural elements, machine learning, and deep learning-based applications and approaches in the context of the Smart grid. In addition in terms of machine learning-based data an-alytics, this paper highlights the limitations of the current research and highlights future directions as well.

While participants in a multi-party multi-turn conversation simultaneously engage in multiple conversation topics, existing response selection methods are developed mainly focusing on a two-party single-conversation scenario. Hence, the prolongation and transition of conversation topics are ignored by current methods. In this work, we frame response selection as a dynamic topic tracking task to match the topic between the response and relevant conversation context. With this new formulation, we propose a novel multi-task learning framework that supports efficient encoding through large pretrained models with only two utterances at once to perform dynamic topic disentanglement and response selection. We also propose Topic-BERT an essential pretraining step to embed topic information into BERT with self-supervised learning. Experimental results on the DSTC-8 Ubuntu IRC dataset show state-of-the-art results in response selection and topic disentanglement tasks outperforming existing methods by a good margin.

Sentence matching is a fundamental task of natural language processing with various applications. Most recent approaches adopt attention-based neural models to build word- or phrase-level alignment between two sentences. However, these models usually ignore the inherent structure within the sentences and fail to consider various dependency relationships among text units. To address these issues, this paper proposes a graph-based approach for sentence matching. First, we represent a sentence pair as a graph with several carefully design strategies. We then employ a novel gated graph attention network to encode the constructed graph for sentence matching. Experimental results demonstrate that our method substantially achieves state-of-the-art performance on two datasets across tasks of natural language and paraphrase identification. Further discussions show that our model can learn meaningful graph structure, indicating its superiority on improved interpretability.

Traditional recommender systems encounter several challenges such as data sparsity and unexplained recommendation. To address these challenges, many works propose to exploit semantic information from review data. However, these methods have two major limitations in terms of the way to model textual features and capture textual interaction. For textual modeling, they simply concatenate all the reviews of a user/item into a single review. However, feature extraction at word/phrase level can violate the meaning of the original reviews. As for textual interaction, they defer the interactions to the prediction layer, making them fail to capture complex correlations between users and items. To address those limitations, we propose a novel Hierarchical Text Interaction model(HTI) for rating prediction. In HTI, we propose to model low-level word semantics and high-level review representations hierarchically. The hierarchy allows us to exploit textual features at different granularities. To further capture complex user-item interactions, we propose to exploit semantic correlations between each user-item pair at different hierarchies. At word level, we propose an attention mechanism specialized to each user-item pair, and capture the important words for representing each review. At review level, we mutually propagate textual features between the user and item, and capture the informative reviews. The aggregated review representations are integrated into a collaborative filtering framework for rating prediction. Experiments on five real-world datasets demonstrate that HTI outperforms state-of-the-art models by a large margin. Further case studies provide a deep insight into HTI's ability to capture semantic correlations at different levels of granularities for rating prediction.

Though it was held this year in October instead of September, Apple's "Hi, Speed" media event was a largely typical iPhone launch party, opening with the expansion of its Siri-powered line of HomePod speakers ("Hi"), and concluding with the long-awaited addition of 5G cellular connectivity to the iPhone lineup ("Speed"). Some companies might have tread cautiously on these topics -- Siri and 5G have both been dogged by complaints -- but Apple didn't hold anything back, using a seemingly endless parade of spokespeople to hype the new devices ahead of preorders. The 5G iPhone 12 family, it promised, will "blast past fast," while the $99 HomePod mini will become a hub to "control your smart home," bringing "intelligent assistant" access to the lowest price yet for any Siri device. Having covered Apple for a long time, I'm not surprised that its latest pitches were all sunshine and roses, but I couldn't help but feel that it was making big promises that could come back to bite the company and its partners. As of October 2020, the only thing less likely to thrill someone than a Siri speaker is typical U.S. 5G network performance, which despite boasts of 1-4Gbps downloads has seen average speeds that are barely better than 4G/LTE. Siri and 5G are both theoretically moving targets -- they're services that could improve at any time and in any region without advance notice -- but prior to this event, neither has delivered on its transformative potential.