The recently proposed Thermodynamic Variational Objective (TVO) leverages thermodynamic integration to provide a family of variational inference objectives, which both tighten and generalize the ubiquitous Evidence Lower Bound (ELBO). However, the tightness of TVO bounds was not previously known, an expensive grid search was used to choose a "schedule" of intermediate distributions, and model learning suffered with ostensibly tighter bounds. In this work, we propose an exponential family interpretation of the geometric mixture curve underlying the TVO and various path sampling methods, which allows us to characterize the gap in TVO likelihood bounds as a sum of KL divergences. We propose to choose intermediate distributions using equal spacing in the moment parameters of our exponential family, which matches grid search performance and allows the schedule to adaptively update over the course of training. Finally, we derive a doubly reparameterized gradient estimator which improves model learning and allows the TVO to benefit from more refined bounds. To further contextualize our contributions, we provide a unified framework for understanding thermodynamic integration and the TVO using Taylor series remainders.
We develop an algorithm for unsupervised domain adaptation (UDA) of a classifier from a labeled source domain to an unlabeled target domain in a sequential learning setting. UDA has been studied extensively recently but the vast majority of the existing methods consider a joint learning setting where the model is trained on the source domain and the target domain data simultaneously. We consider a more practical setting, where the model has been trained on the labeled source domain data and then needs to be adapted to the unlabeled source domain, without having access to the source domain training data. We tackle this problem by aligning the distributions of the source and the target domain in a discriminative embedding space. To overcome the challenges of learning in a sequential setting, we learn an intermediate prototypical distribution from the source labeled data and then use this distribution for knowledge transfer to the target domain. We provide theoretical justification for the proposed algorithm by showing that it optimizes an upper-bound for the expected risk in the target domain. We also conduct extensive experiments with several standard benchmarks and demonstrate the competitiveness of the proposed method compared to existing joint learning UDA algorithms.
When specific events seem to spur others in their wake, marked Hawkes processes enable us to reckon with their statistics. The underdetermined empirical nature of these event-triggering mechanisms hinders estimation in the multivariate setting. Spatiotemporal applications alleviate this obstacle by allowing relationships to depend only on relative distances in real Euclidean space; we employ the framework as a vessel for embedding arbitrary event types in a new latent space. By performing synthetic experiments on short records as well as an investigation into options markets and pathogens, we demonstrate that learning the embedding alongside a point process model uncovers the coherent, rather than spurious, interactions.
Many important phenomena arise naturally as temporal point processes with different types of events influencing future events in complex ways. Estimation of multivariate point processes is a notorious proposition. We take inspiration from spatiotemporal point processes, where relationships depend only on relative distances in real Euclidean space, to suggest embedding arbitrary event types in a latent space. We demonstrate that we can simultaneously learn this embedding and a point process model to recover relationships among events.
In the presence of noisy or incorrect labels, neural networks have the undesirable tendency to memorize information about the noise. Standard regularization techniques such as dropout, weight decay or data augmentation sometimes help, but do not prevent this behavior. If one considers neural network weights as random variables that depend on the data and stochasticity of training, the amount of memorized information can be quantified with the Shannon mutual information between weights and the vector of all training labels given inputs, $I(w : \mathbf{y} \mid \mathbf{x})$. We show that for any training algorithm, low values of this term correspond to reduction in memorization of label-noise and better generalization bounds. To obtain these low values, we propose training algorithms that employ an auxiliary network that predicts gradients in the final layers of a classifier without accessing labels. We illustrate the effectiveness of our approach on versions of MNIST, CIFAR-10, and CIFAR-100 corrupted with various noise models, and on a large-scale dataset Clothing1M that has noisy labels.
Tackling binary analysis problems has traditionally implied manually defining rules and heuristics. As an alternative, we are suggesting using machine learning models for learning distributed representations of binaries that can be applicable for a number of downstream tasks. We construct a computational graph from the binary executable and use it with a graph convolutional neural network to learn a high dimensional representation of the program. We show the versatility of this approach by using our representations to solve two semantically different binary analysis tasks -- algorithm classification and vulnerability discovery. We compare the proposed approach to our own strong baseline as well as published results and demonstrate improvement on the state of the art methods for both tasks.
User engagement is a critical metric for evaluating the quality of open-domain dialogue systems. Prior work has focused on conversation-level engagement by using heuristically constructed features such as the number of turns and the total time of the conversation. In this paper, we investigate the possibility and efficacy of estimating utterance-level engagement and define a novel metric, {\em predictive engagement}, for automatic evaluation of open-domain dialogue systems. Our experiments demonstrate that (1) human annotators have high agreement on assessing utterance-level engagement scores; (2) conversation-level engagement scores can be predicted from properly aggregated utterance-level engagement scores. Furthermore, we show that the utterance-level engagement scores can be learned from data. These scores can improve automatic evaluation metrics for open-domain dialogue systems, as shown by correlation with human judgements. This suggests that predictive engagement can be used as a real-time feedback for training better dialogue models.
We study the bias in several state-of-the-art named entity recognition (NER) models---specifically, a difference in the ability to recognize male and female names as PERSON entity types. We evaluate NER models on a dataset containing 139 years of U.S. census baby names and find that relatively more female names, as opposed to male names, are not recognized as PERSON entities. We study the extent of this bias in several NER systems that are used prominently in industry and academia. In addition, we also report a bias in the datasets on which these models were trained. The result of this analysis yields a new benchmark for gender bias evaluation in named entity recognition systems. The data and code for the application of this benchmark will be publicly available for researchers to use.
We propose a novel deep structured learning framework for event temporal relation extraction. The model consists of 1) a recurrent neural network (RNN) to learn scoring functions for pair-wise relations, and 2) a structured support vector machine (SSVM) to make joint predictions. The neural network automatically learns representations that account for long-term contexts to provide robust features for the structured model, while the SSVM incorporates domain knowledge such as transitive closure of temporal relations as constraints to make better globally consistent decisions. By jointly training the two components, our model combines the benefits of both data-driven learning and knowledge exploitation. Experimental results on three high-quality event temporal relation datasets (TCR, MATRES, and TB-Dense) demonstrate that incorporated with pre-trained contextualized embeddings, the proposed model achieves significantly better performances than the state-of-the-art methods on all three datasets. We also provide thorough ablation studies to investigate our model.