
Regularizing Blackbox Models for Improved Interpretability
Most work on interpretability in machine learning has focused on designing either inherently interpretable models, that typically tradeoff interpretability for accuracy, or posthoc explanation systems, that lack guarantees about their explanation quality. We propose an alternative to these approaches by directly regularizing a blackbox model for interpretability at training time. Our approach explicitly connects three key aspects of interpretable machine learning: the model's innate explainability, the explanation system used at test time, and the metrics that measure explanation quality. Our regularization results in substantial (up to orders of magnitude) improvement in terms of explanation fidelity and stability metrics across a range of datasets, models, and blackbox explanation systems. Remarkably, our regularizers also slightly improve predictive accuracy on average across the nine datasets we consider. Further, we show that the benefits of our novel regularizers on explanation quality provably generalize to unseen test points.
02/18/2019 ∙ by Gregory Plumb, et al. ∙ 20 ∙ shareread it

Consistency by Agreement in Zeroshot Neural Machine Translation
Generalization and reliability of multilingual translation often highly depend on the amount of available parallel data for each language pair of interest. In this paper, we focus on zeroshot generalizationa challenging setup that tests models on translation directions they have not been optimized for at training time. To solve the problem, we (i) reformulate multilingual translation as probabilistic inference, (ii) define the notion of zeroshot consistency and show why standard training often results in models unsuitable for zeroshot tasks, and (iii) introduce a consistent agreementbased training method that encourages the model to produce equivalent translations of parallel sentences in auxiliary languages. We test our multilingual NMT models on multiple public zeroshot translation benchmarks (IWSLT17, UN corpus, Europarl) and show that agreementbased learning often results in 23 BLEU zeroshot improvement over strong baselines without any loss in performance on supervised translation directions.
04/04/2019 ∙ by Maruan AlShedivat, et al. ∙ 16 ∙ shareread it

On the Complexity of Exploration in GoalDriven Navigation
Building agents that can explore their environments intelligently is a challenging open problem. In this paper, we make a step towards understanding how a hierarchical design of the agent's policy can affect its exploration capabilities. First, we design EscapeRoom environments, where the agent must figure out how to navigate to the exit by accomplishing a number of intermediate tasks (subgoals), such as finding keys or opening doors. Our environments are procedurally generated and vary in complexity, which can be controlled by the number of subgoals and relationships between them. Next, we propose to measure the complexity of each environment by constructing dependency graphs between the goals and analytically computing hitting times of a random walk in the graph. We empirically evaluate Proximal Policy Optimization (PPO) with sparse and shaped rewards, a variation of policy sketches, and a hierarchical version of PPO (called HiPPO) akin to hDQN. We show that analytically estimated hitting time in goal dependency graphs is an informative metric of the environment complexity. We conjecture that the result should hold for environments other than navigation. Finally, we show that solving environments beyond certain level of complexity requires hierarchical approaches.
11/16/2018 ∙ by Maruan AlShedivat, et al. ∙ 10 ∙ shareread it

Regularizing Blackbox Models for Improved Interpretability (HILL 2019 Version)
Most of the work on interpretable machine learning has focused on designing either inherently interpretable models, which typically tradeoff accuracy for interpretability, or posthoc explanation systems, which lack guarantees about their explanation quality. We propose an alternative to these approaches by directly regularizing a blackbox model for interpretability at training time. Our approach explicitly connects three key aspects of interpretable machine learning: (i) the model's innate explainability, (ii) the explanation system used at test time, and (iii) the metrics that measure explanation quality. Our regularization results in substantial improvement in terms of the explanation fidelity and stability metrics across a range of datasets and blackbox explanation systems while slightly improving accuracy. Further, if the resulting model is still not sufficiently interpretable, the weight of the regularization term can be adjusted to achieve the desired tradeoff between accuracy and interpretability. Finally, we justify theoretically that the benefits of explanationbased regularization generalize to unseen points.
05/31/2019 ∙ by Gregory Plumb, et al. ∙ 5 ∙ shareread it

Continuous Adaptation via MetaLearning in Nonstationary and Competitive Environments
Ability to continuously learn and adapt from limited experience in nonstationary environments is an important milestone on the path towards general intelligence. In this paper, we cast the problem of continuous adaptation into the learningtolearn framework. We develop a simple gradientbased metalearning algorithm suitable for adaptation in dynamically changing and adversarial scenarios. Additionally, we design a new multiagent competitive environment, RoboSumo, and define iterated adaptation games for testing various aspects of continuous adaptation strategies. We demonstrate that metalearning enables significantly more efficient adaptation than reactive baselines in the fewshot regime. Our experiments with a population of agents that learn and compete suggest that metalearners are the fittest.
10/10/2017 ∙ by Maruan AlShedivat, et al. ∙ 0 ∙ shareread it

Stochastic Synapses Enable Efficient BrainInspired Learning Machines
Recent studies have shown that synaptic unreliability is a robust and sufficient mechanism for inducing the stochasticity observed in cortex. Here, we introduce Synaptic Sampling Machines, a class of neural network models that uses synaptic stochasticity as a means to Monte Carlo sampling and unsupervised learning. Similar to the original formulation of Boltzmann machines, these models can be viewed as a stochastic counterpart of Hopfield networks, but where stochasticity is induced by a random mask over the connections. Synaptic stochasticity plays the dual role of an efficient mechanism for sampling, and a regularizer during learning akin to DropConnect. A local synaptic plasticity rule implementing an eventdriven form of contrastive divergence enables the learning of generative models in an online fashion. Synaptic sampling machines perform equally well using discretetimed artificial units (as in Hopfield networks) or continuoustimed leaky integrate & fire neurons. The learned representations are remarkably sparse and robust to reductions in bit precision and synapse pruning: removal of more than 75 followed by cursory relearning causes a negligible performance loss on benchmark classification tasks. The spiking neuronbased synaptic sampling machines outperform existing spikebased unsupervised learners, while potentially offering substantial advantages in terms of power and complexity, and are thus promising models for online learning in braininspired hardware.
11/14/2015 ∙ by Emre O. Neftci, et al. ∙ 0 ∙ shareread it

Contextual Explanation Networks
We introduce contextual explanation networks (CENs)a class of models that learn to predict by generating and leveraging intermediate explanations. CENs combine deep networks with contextspecific probabilistic models and construct explanations in the form of locallycorrect hypotheses. Contrary to the existing posthoc modelexplanation tools, CENs learn to predict and to explain jointly. Our approach offers two major advantages: (i) for each prediction, valid instancespecific explanations are generated with no computational overhead and (ii) prediction via explanation acts as a regularization and boosts performance in lowresource settings. We prove that local approximations to the decision boundary of our networks are consistent with the generated explanations. Our results on image and text classification and survival analysis tasks demonstrate that CENs can easily match or outperform the stateoftheart while offering additional insights behind each prediction, valuable for decision support.
05/29/2017 ∙ by Maruan AlShedivat, et al. ∙ 0 ∙ shareread it

Learning Nondeterministic Representations with Energybased Ensembles
The goal of a generative model is to capture the distribution underlying the data, typically through latent variables. After training, these variables are often used as a new representation, more effective than the original features in a variety of learning tasks. However, the representations constructed by contemporary generative models are usually pointwise deterministic mappings from the original feature space. Thus, even with representations robust to classspecific transformations, statistically driven models trained on them would not be able to generalize when the labeled data is scarce. Inspired by the stochasticity of the synaptic connections in the brain, we introduce Energybased Stochastic Ensembles. These ensembles can learn nondeterministic representations, i.e., mappings from the feature space to a family of distributions in the latent space. These mappings are encoded in a distribution over a (possibly infinite) collection of models. By conditionally sampling models from the ensemble, we obtain multiple representations for every input example and effectively augment the data. We propose an algorithm similar to contrastive divergence for training restricted Boltzmann stochastic ensembles. Finally, we demonstrate the concept of the stochastic representations on a synthetic dataset as well as test them in the oneshot learning scenario on MNIST.
12/23/2014 ∙ by Maruan AlShedivat, et al. ∙ 0 ∙ shareread it

Learning Scalable Deep Kernels with Recurrent Structure
Many applications in speech, robotics, finance, and biology deal with sequential data, where ordering matters and recurrent structures are common. However, this structure cannot be easily captured by standard kernel functions. To model such structure, we propose expressive closedform kernel functions for Gaussian processes. The resulting model, GPLSTM, fully encapsulates the inductive biases of long shortterm memory (LSTM) recurrent networks, while retaining the nonparametric probabilistic advantages of Gaussian processes. We learn the properties of the proposed kernels by optimizing the Gaussian process marginal likelihood using a new provably convergent semistochastic gradient procedure and exploit the structure of these kernels for scalable training and prediction. This approach provides a practical representation for Bayesian LSTMs. We demonstrate stateoftheart performance on several benchmarks, and thoroughly investigate a consequential autonomous driving application, where the predictive uncertainties provided by GPLSTM are uniquely valuable.
10/27/2016 ∙ by Maruan AlShedivat, et al. ∙ 0 ∙ shareread it

Learning HMMs with Nonparametric Emissions via Spectral Decompositions of Continuous Matrices
Recently, there has been a surge of interest in using spectral methods for estimating latent variable models. However, it is usually assumed that the distribution of the observations conditioned on the latent variables is either discrete or belongs to a parametric family. In this paper, we study the estimation of an mstate hidden Markov model (HMM) with only smoothness assumptions, such as Hölderian conditions, on the emission densities. By leveraging some recent advances in continuous linear algebra and numerical analysis, we develop a computationally efficient spectral algorithm for learning nonparametric HMMs. Our technique is based on computing an SVD on nonparametric estimates of density functions by viewing them as continuous matrices. We derive sample complexity bounds via concentration results for nonparametric density estimation and novel perturbation theory results for continuous matrices. We implement our method using Chebyshev polynomial approximations. Our method is competitive with other baselines on synthetic and real problems and is also very computationally efficient.
09/21/2016 ∙ by Kirthevasan Kandasamy, et al. ∙ 0 ∙ shareread it

Personalized Survival Prediction with Contextual Explanation Networks
Accurate and transparent prediction of cancer survival times on the level of individual patients can inform and improve patient care and treatment practices. In this paper, we design a model that concurrently learns to accurately predict patientspecific survival distributions and to explain its predictions in terms of patient attributes such as clinical tests or assessments. Our model is flexible and based on a recurrent network, can handle various modalities of data including temporal measurements, and yet constructs and uses simple explanations in the form of patient and timespecific linear regression. For analysis, we use two publicly available datasets and show that our networks outperform a number of baselines in prediction while providing a way to inspect the reasons behind each prediction.
01/30/2018 ∙ by Maruan AlShedivat, et al. ∙ 0 ∙ shareread it
Maruan AlShedivat
is this you? claim profile
Ph.D. student in Machine Learning at Carnegie Mellon University