ICML 2017 论文列表

Developing a dialogue agent that is capable of making autonomous decisions and communicating by natural language is one of the long-term goals of machine learning research. The traditional approaches either rely on hand-crafting a small state-action set for applying reinforcement learning that is not scalable or constructing deterministic models for learning dialogue sentences that fail to capture the conversational stochasticity. In this paper, however, we propose a Latent Intention Dialogue Model that employs a discrete latent variable to learn underlying dialogue intentions in the framework of neural variational inference. Additionally, in a goal-oriented dialogue scenario, the latent intentions can be interpreted as actions guiding the generation of machine responses, which can be further refined autonomously by reinforcement learning. The experiments demonstrate the effectiveness of discrete latent variable models on learning goal-oriented dialogues, and the results outperform the published benchmarks on both corpus-based evaluation and human evaluation.

The latent feature model (LFM), proposed in (Griffiths \& Ghahramani, 2005), but possibly with earlier origins, is a generalization of a mixture model, where each instance is generated not from a single latent class but from a combination of

Neural networks have proven effective at solving difficult problems but designing their architectures can be challenging, even for image classification problems alone. Our goal is to minimize human participation, so we employ evolutionary algorithms to discover such networks automatically. Despite significant computational requirements, we show that it is now possible to evolve models with accuracies within the range of those published in the last year. Specifically, we employ simple evolutionary techniques at unprecedented scales to discover models for the CIFAR-10 and CIFAR-100 datasets, starting from trivial initial conditions and reaching accuracies of 94.6\% (95.6\% for ensemble) and 77.0\%, respectively. To do this, we use novel and intuitive mutation operators that navigate large search spaces; we stress that no human participation is required once evolution starts and that the output is a fully-trained model. Throughout this work, we place special emphasis on the repeatability of results, the variability in the outcomes and the computational requirements.

The pre-dominant approach to language modeling to date is based on recurrent neural networks. Their success on this task is often linked to their ability to capture unbounded context. In this paper we develop a finite context approach through stacked convolutions, which can be more efficient since they allow parallelization over sequential tokens. We propose a novel simplified gating mechanism that outperforms Oord et al. (2016) and investigate the impact of key architectural decisions. The proposed approach achieves state-of-the-art on the WikiText-103 benchmark, even though it features long-term dependencies, as well as competitive results on the Google Billion Words benchmark. Our model reduces the latency to score a sentence by an order of magnitude compared to a recurrent baseline. To our knowledge, this is the first time a non-recurrent approach is competitive with strong recurrent models on these large scale language tasks.

The availability of large scale event data with time stamps has given rise to dynamically evolving knowledge graphs that contain temporal information for each edge. Reasoning over time in such dynamic knowledge graphs is not yet well understood. To this end, we present Know-Evolve, a novel deep evolutionary knowledge network that learns non-linearly evolving entity representations over time. The occurrence of a fact (edge) is modeled as a multivariate point process whose intensity function is modulated by the score for that fact computed based on the learned entity embeddings. We demonstrate significantly improved performance over various relational learning approaches on two large scale real-world datasets. Further, our method effectively predicts occurrence or recurrence time of a fact which is novel compared to prior reasoning approaches in multi-relational setting.

In the context of supervised tensor learning, preserving the structural information and exploiting the discriminative nonlinear relationships of tensor data are crucial for improving the performance of learning tasks. Based on tensor factorization theory and kernel methods, we propose a novel Kernelized Support Tensor Machine (KSTM) which integrates kernelized tensor factorization with maximum-margin criterion. Specifically, the kernelized factorization technique is introduced to approximate the tensor data in kernel space such that the complex nonlinear relationships within tensor data can be explored. Further, dual structural preserving kernels are devised to learn the nonlinear boundary between tensor data. As a result of joint optimization, the kernels obtained in KSTM exhibit better generalization power to discriminative analysis. The experimental results on real-world neuroimaging datasets show the superiority of KSTM over the state-of-the-art techniques.

We address the problem of learning a ranking by using adaptively chosen pairwise comparisons. Our goal is to recover the ranking accurately but to sample the comparisons sparingly. If all comparison outcomes are consistent with the ranking, the optimal solution is to use an efficient sorting algorithm, such as Quicksort. But how do sorting algorithms behave if some comparison outcomes are inconsistent with the ranking? We give favorable guarantees for Quicksort for the popular Bradley-Terry model, under natural assumptions on the parameters. Furthermore, we empirically demonstrate that sorting algorithms lead to a very simple and effective active learning strategy: repeatedly sort the items. This strategy performs as well as state-of-the-art methods (and much better than random sampling) at a minuscule fraction of the computational cost.

To be tractable and robust to data noise, existing metric learning algorithms commonly rely on PCA as a pre-processing step. How can we know, however, that PCA, or any other specific dimensionality reduction technique, is the method of choice for the problem at hand? The answer is simple: We cannot! To address this issue, in this paper, we develop a Riemannian framework to jointly learn a mapping performing dimensionality reduction and a metric in the induced space. Our experiments evidence that, while we directly work on high-dimensional features, our approach yields competitive runtimes with and higher accuracy than state-of-the-art metric learning algorithms.

In this paper, we consider the problem of machine teaching, the inverse problem of machine learning. Different from traditional machine teaching which views the learners as batch algorithms, we study a new paradigm where the learner uses an iterative algorithm and a teacher can feed examples sequentially and intelligently based on the current performance of the learner. We show that the teaching complexity in the iterative case is very different from that in the batch case. Instead of constructing a minimal training set for learners, our iterative machine teaching focuses on achieving fast convergence in the learner model. Depending on the level of information the teacher has from the learner model, we design teaching algorithms which can provably reduce the number of teaching examples and achieve faster convergence than learning without teachers. We also validate our theoretical findings with extensive experiments on different data distribution and real image datasets.

This paper investigates the problem of interactively learning behaviors communicated by a human teacher using positive and negative feedback. Much previous work on this problem has made the assumption that people provide feedback for decisions that is dependent on the behavior they are teaching and is independent from the learner’s current policy. We present empirical results that show this assumption to be false—whether human trainers give a positive or negative feedback for a decision is influenced by the learner’s current policy. Based on this insight, we introduce Convergent Actor-Critic by Humans (COACH), an algorithm for learning from policy-dependent feedback that converges to a local optimum. Finally, we demonstrate that COACH can successfully learn multiple behaviors on a physical robot.

There exist many problem domains where the interpretability of neural network models is essential for deployment. Here we introduce a recurrent architecture composed of input-switched affine transformations – in other words an RNN without any explicit nonlinearities, but with input-dependent recurrent weights. This simple form allows the RNN to be analyzed via straightforward linear methods: we can exactly characterize the linear contribution of each input to the model predictions; we can use a change-of-basis to disentangle input, output, and computational hidden unit subspaces; we can fully reverse-engineer the architecture’s solution to a simple task. Despite this ease of interpretation, the input switched affine network achieves reasonable performance on a text modeling tasks, and allows greater computational efficiency than networks with standard nonlinearities.

This paper presents the input convex neural network architecture. These are scalar-valued (potentially deep) neural networks with constraints on the network parameters such that the output of the network is a convex function of (some of) the inputs. The networks allow for efficient inference via optimization over some inputs to the network given others, and can be applied to settings including structured prediction, data imputation, reinforcement learning, and others. In this paper we lay the basic groundwork for these models, proposing methods for inference, optimization and learning, and analyze their representational power. We show that many existing neural network architectures can be made input-convex with a minor modification, and develop specialized optimization algorithms tailored to this setting. Finally, we highlight the performance of the methods on multi-label prediction, image completion, and reinforcement learning problems, where we show improvement over the existing state of the art in many cases.

This paper presents a new scalable approach, termed Innovation Pursuit (iPursuit), to the problem of subspace clustering. iPursuit rests on a new geometrical idea whereby each subspace is identified based on its novelty with respect to the other subspaces. The subspaces are identified consecutively by solving a series of simple linear optimization problems, each searching for a direction of innovation in the span of the data. A detailed mathematical analysis is provided establishing sufficient conditions for the proposed approach to correctly cluster the data points. Moreover, the proposed direction search approach can be integrated with spectral clustering to yield a new variant of spectral-clustering-based algorithms. Remarkably, the proposed approach can provably yield exact clustering even when the subspaces have significant intersections. The numerical simulations demonstrate that iPursuit can often outperform the state-of-the-art subspace clustering algorithms – more so for subspaces with significant intersections – along with substantial reductions in computational complexity.

The classic algorithm of Viterbi computes the most likely path in a Hidden Markov Model (HMM) that results in a given sequence of observations. It runs in time $O(Tn^2)$ given a sequence of T observations from a HMM with n states. Despite significant interest in the problem and prolonged effort by different communities, no known algorithm achieves more than a polylogarithmic speedup. In this paper, we explain this difficulty by providing matching conditional lower bounds. Our lower bounds are based on assumptions that the best known algorithms for the All-Pairs Shortest Paths problem (APSP) and for the Max-Weight k-Clique problem in edge-weighted graphs are essentially tight. Finally, using a recent algorithm by Green Larsen and Williams for online Boolean matrix-vector multiplication, we get a $2^{\Omega(\sqrt{\log n})}$ speedup for the Viterbi algorithm when there are few distinct transition probabilities in the HMM.

Recently, reinforcement learning with deep neural networks has achieved great success in challenging continuous control problems such as 3D locomotion and robotic manipulation. However, in real-world control problems, the actions one can take are bounded by physical constraints, which introduces a bias when the standard Gaussian distribution is used as the stochastic policy. In this work, we propose to use the Beta distribution as an alternative and analyze the bias and variance of the policy gradients of both policies. We show that the Beta policy is bias-free and provides significantly faster convergence and higher scores over the Gaussian policy when both are used with trust region policy optimization (TRPO) and actor critic with experience replay (ACER), the state-of-the-art on- and off-policy stochastic methods respectively, on OpenAI Gym’s and MuJoCo’s continuous control environments.

The pairwise influence matrix of Dobrushin has long been used as an analytical tool to bound the rate of convergence of Gibbs sampling. In this work, we use Dobrushin influence as the basis of a practical tool to certify and efficiently improve the quality of a Gibbs sampler. Our Dobrushin-optimized Gibbs samplers (DoGS) offer customized variable selection orders for a given sampling budget and variable subset of interest, explicit bounds on total variation distance to stationarity, and certifiable improvements over the standard systematic and uniform random scan Gibbs samplers. In our experiments with image segmentation, Markov chain Monte Carlo maximum likelihood estimation, and Ising model inference, DoGS consistently deliver higher-quality inferences with significantly smaller sampling budgets than standard Gibbs samplers.

Recent work on generative text modeling has found that variational autoencoders (VAE) with LSTM decoders perform worse than simpler LSTM language models (Bowman et al., 2015). This negative result is so far poorly understood, but has been attributed to the propensity of LSTM decoders to ignore conditioning information from the encoder. In this paper, we experiment with a new type of decoder for VAE: a dilated CNN. By changing the decoder’s dilation architecture, we control the size of context from previously generated words. In experiments, we find that there is a trade-off between contextual capacity of the decoder and effective use of encoding information. We show that when carefully managed, VAEs can outperform LSTM language models. We demonstrate perplexity gains on two datasets, representing the first positive language modeling result with VAE. Further, we conduct an in-depth investigation of the use of VAE (with our new decoding architecture) for semi-supervised and unsupervised labeling tasks, demonstrating gains over several strong baselines.

We present a neural encoder-decoder model to convert images into presentational markup based on a scalable coarse-to-fine attention mechanism. Our method is evaluated in the context of image-to-LaTeX generation, and we introduce a new dataset of real-world rendered mathematical expressions paired with LaTeX markup. We show that unlike neural OCR techniques using CTC-based models, attention-based approaches can tackle this non-standard OCR task. Our approach outperforms classical mathematical OCR systems by a large margin on in-domain rendered data, and, with pretraining, also performs well on out-of-domain handwritten data. To reduce the inference complexity associated with the attention-based approaches, we introduce a new coarse-to-fine attention layer that selects a support region before applying attention.

Motivated by applications in computational advertising and systems biology, we consider the problem of identifying the best out of several possible soft interventions at a source node $V$ in an acyclic causal directed graph, to maximize the expected value of a target node $Y$ (located downstream of $V$). Our setting imposes a fixed total budget for sampling under various interventions, along with cost constraints on different types of interventions. We pose this as a best arm identification bandit problem with $K$ arms, where each arm is a soft intervention at $V$ and leverage the information leakage among the arms to provide the first gap dependent error and simple regret bounds for this problem. Our results are a significant improvement over the traditional best arm identification results. We empirically show that our algorithms outperform the state of the art in the Flow Cytometry data-set, and also apply our algorithm for model interpretation of the Inception-v3 deep net that classifies images.