Publications

We study offline policy evaluation in a setting where the target policy can take actions that were not available when the data was logged. We analyze the bias of two popular regression-based estimators in this setting, and upper-bound their biases by a quantity we refer to as the reward regression risk. We show that the estimators can be asymptotically unbiased and uniformly convergent if the reward regression

We tackle the problem of predicting the number of optimization steps that a pretrained deep network needs to converge to a given value of the loss function. To do so, we leverage the fact that the training dynamics of a deep network during fine-tuning are well approximated by those of a linearized model. This allows us to approximate the training loss and accuracy at any point during training by solving

Neural network based approaches to data-to-text natural language generation (NLG) have gained popularity in recent years, with the goal of generating a natural language prompt that accurately realizes an input meaning representation. To facilitate the training of neural network models, researchers created large datasets of paired utterances and their meaning representations. However, the creation of such

Current voice assistants typically use the best hypothesis yielded by their Automatic Speech Recognition (ASR) module as input to their Natural Language Understanding (NLU) module, thereby losing helpful information that might be stored in lower-ranked ASR hypotheses. We explore the change in performance of NLU associated tasks when utilizing fivebest ASR hypotheses when compared to status quo for two language

When scaling distributed training, the communication overhead is often the bottleneck. In this paper, we propose a novel SGD variant with reduced communication and adaptive learning rates. We prove the convergence of the proposed algorithm for smooth but non-convex problems. Empirical results show that the proposed algorithm significantly reduces the communication overhead, which, in turn, reduces the training

Machine learning is an important tool for analyzing large data sets, but its use on sensitive data may be limited by regulation. One solution to this problem is to perform machine learning tasks on encrypted data using homomorphic encryption, which enables arbitrary computation on encrypted data. We take a fresh look at one specific task: training a logistic regression model on encrypted data. The most

Goal-oriented dialog systems enable users to complete specific goals like requesting information about a movie or booking a ticket. Typically the dialog system pipeline contains multiple ML models, including natural language understanding, state tracking and action prediction (policy learning). These models are trained through a combination of supervised or reinforcement learning methods and therefore require

We study privacy in federated learning systems where the model is partitioned into global and local components, the latter of which are personalized for each participating client and never shared. This setting suggests a new type of privacy breach: that the server might learn a client’s local model from updates to the global model. Using on-device recommendation as a motivating example, we show that this

Although machine learning (ML) is widely used for predictive tasks, there are important scenarios in which ML cannot be used or at least cannot achieve its full potential. A major barrier to adoption is the sensitive nature of predictive queries. Individual users may lack sufficiently rich datasets to train accurate models locally but also be unwilling to send sensitive queries to commercial services that

Automated machine learning (AutoML) can produce complex model ensembles by stacking, bagging, and boosting many individual models like trees, deep networks, and nearest neighbor estimators. While highly accurate, the resulting predictors are large, slow, and opaque as compared to their constituents. To improve the deployment of AutoML on tabular data, we propose FAST-DAD to distill arbitrarily-complex ensemble

Developing machine learning algorithms to understand personto-person engagement can result in natural user experiences for communal devices such as Amazon Alexa. Among other cues such as voice activity and gaze, a person’s audio-visual expression that includes tone of the voice and facial expression serves as an implicit signal of engagement between parties in a dialog. This study investigates deep-learning

In a hybrid key encapsulation construction, multiple independent key encapsulation mechanisms are combined in a way that ensures the resulting key is secure according to the strongest mechanism. Such constructions can combine mechanisms that are secure in different settings and achieve the combined security of all mechanisms. For example classical and post-quantum mechanisms can be combined in order to

Deep Reinforcement Learning (RL) has demonstrated to be useful for a wide variety of robotics applications. To address sample efficiency and safety during training, it is common to train Deep RL policies in a simulator and then deploy to the real world, a process called Sim2Real transfer. For robotics applications, the deployment heterogeneities and runtime compute stochasticity results in variable timing

The COVID-19 pandemic is the worst pandemic to strike the world in over a century. Crucial to stemming the tide of the SARSCoV-2 virus is communicating to vulnerable populations the means by which they can protect themselves. To this end, the collaborators forming the Translation Initiative for COVID-19 (TICO-19)1 have made test and development data available to AI and MT researchers in 35 different languages

Online recommendation is an essential functionality across a variety of services, including e-commerce and video streaming, where items to buy, watch, or read are suggested to users. Justifying recommendations, i.e., explaining why a user might like the recommended item, has been shown to improve user satisfaction and persuasiveness of the recommendation. In this paper, we develop a method for generating