211 research outputs found
Evaluating the accuracy of data collection on mobile phones: A study of forms, SMS, and voice
No full textWhile mobile phones have found broad application in reporting health, financial, and environmental data, there has been little study of the possible errors incurred during mobile data collection. This paper provides the first (to our knowledge) quantitative evaluation of data entry accuracy on mobile phones in a resource-poor setting. Via a study of 13 users in Gujarat, India, we evaluated three user interfaces: 1) electronic forms, containing numeric fields and multiple-choice menus, 2) SMS, where users enter delimited text messages according to printed cue cards, and 3) voice, where users call an operator and dictate the data in real-time. Our results indicate error rates (per datum entered) of 4.2% for electronic forms, 4.8% for SMS, and 0.45% for voice. These results caused us to migrate our own initiative (a tuberculosis treatment program in rural India) from electronic forms to voice, in order to avoid errors on critical health data. While our study has some limitations, including varied backgrounds and training of participants, it suggests that some care is needed in deploying electronic interfaces in resource-poor settings. Further, it raises the possibility of using voice as a low-tech, high-accuracy, and cost-effective interface for mobile data collection.Massachusetts Institute of Technology. Public Service Cente
Human-Focused Reinforcement Learning
No full textPresented on February 27, 2020 at 12:00 p.m. in the Technology Square Research Building, Banquet Hall.Emma Brunskill is an assistant professor in the Computer Science Department at Stanford University where she leads the AI for Human Impact group. Her work focuses on reinforcement learning in high stakes scenarios--how can an agent learn from experience to make good decisions when experience is costly or risky, such as in educational software, healthcare decision making, or people-facing applications.Runtime: 56:32 minutesThere is increasing excitement about reinforcement learning--a subarea of machine learning for enabling an agent to learn to make good decisions. Yet numerous questions and challenges remain for reinforcement learning to help support progress in applications that involve interacting with people, like education, consumer marketing and healthcare. I will discuss our work on some of the technical challenges that arise in this pursuit, including sample efficiency, counterfactual reasoning, robustness, and applications to health and education
Where to go: Interpreting natural directions using global inference
No full textAn important component of human-robot interaction is that people need to be able to instruct robots to move to other locations using naturally given directions. When giving directions, people often make mistakes such as labelling errors (e.g., left vs. right) and errors of omission (skipping important decision points in a sequence). Furthermore, people often use multiple levels of granularity in specifying directions, referring to locations using single object landmarks, multiple landmarks in a given location, or identifying large regions as a single location. The challenge is to identify the correct path to a destination from a sequence of noisy, possibly erroneous directions. In our work we cast this problem as probabilistic inference: given a set of directions, an agent should automatically find the path with the geometry and physical appearance to maximize the likelihood of those directions. We use a specific variant of a Markov Random Field (MRF) to represent our model, and gather multi-granularity representation information using existing large tagged datasets. On a dataset of route directions collected in a large third floor university building, we found that our algorithm correctly inferred the true final destination in 47 out of the 55 cases successfully followed by humans volunteers. These results suggest that our algorithm is performing well relative to human users. In the future this work will be included in a broader system for autonomously constructing environmental representations that support natural human-robot interaction for direction giving.United States. Air Force Office of Scientific Research (Agile Robotics project, contract number 7000038334)National Science Foundation (U.S.) (NSF Division of Information and Intelligent Systems under grant # 0546467)Massachusetts Institute of Technology (Hugh Hampton Young Memorial Fund Fellowship)United States. Office of Naval Research (MURI N00014-07-1-0749
BNN-DP: Robustness Certification of Bayesian Neural Networks via Dynamic Programming
No full textIn this paper, we introduce BNN-DP, an efficient algorithmic framework for analysis of adversarial robustness of Bayesian Neural Networks (BNNs). Given a compact set of input points T ⊂ Rn, BNN-DP computes lower and upper bounds on the BNN's predictions for all the points in T. The framework is based on an interpretation of BNNs as stochastic dynamical systems, which enables the use of Dynamic Programming (DP) algorithms to bound the prediction range along the layers of the network. Specifically, the method uses bound propagation techniques and convex relaxations to derive a backward recursion procedure to over-approximate the prediction range of the BNN with piecewise affine functions. The algorithm is general and can handle both regression and classification tasks. On a set of experiments on various regression and classification tasks and BNN architectures, we show that BNN-DP outperforms state-of-the-art methods by up to four orders of magnitude in both tightness of the bounds and computational efficiency.Team Luca Laurent
On Second-Order Scoring Rules for Epistemic Uncertainty Quantification
No full textIt is well known that accurate probabilistic predictors can be trained through empirical risk minimisation with proper scoring rules as loss functions. While such learners capture so-called aleatoric uncertainty of predictions, various machine learning methods have recently been developed with the goal to let the learner also represent its epistemic uncertainty, i.e., the uncertainty caused by a lack of knowledge and data. An emerging branch of the literature proposes the use of a second-order learner that provides predictions in terms of distributions on probability distributions. However, recent work has revealed serious theoretical shortcomings for second-order predictors based on loss minimisation. In this paper, we generalise these findings and prove a more fundamental result: There seems to be no loss function that provides an incentive for a second-order learner to faithfully represent its epistemic uncertainty in the same manner as proper scoring rules do for standard (first-order) learners. As a main mathematical tool to prove this result, we introduce the generalised notion of second-order scoring rules
Computational Asymmetries in Robust Classification
No full textIn the context of adversarial robustness, we make three strongly related contributions. First, we prove that while attacking ReLU classifiers is -hard, ensuring their robustness at training time is -hard (even on a single example). This asymmetry provides a rationale for the fact that robust classifications approaches are frequently fooled in the literature. Second, we show that inference-time robustness certificates are not affected by this asymmetry, by introducing a proof-of-concept approach named Counter-Attack (CA). Indeed, CA displays a reversed asymmetry: running the defense is -hard, while attacking it is -hard. Finally, motivated by our previous result, we argue that adversarial attacks can be used in the context of robustness certification, and provide an empirical evaluation of their effectiveness. As a byproduct of this process, we also release UG100, a benchmark dataset for adversarial attacks
Global optimality of Elman-type RNNs in the mean-field regime
No full textWe analyze Elman-type recurrent neural networks (RNNs) and their training in the mean-field regime. Specifically, we show convergence of gradient descent training dynamics of the RNN to the corresponding mean-field formulation in the large width limit. We also show that the fixed points of the limiting infinite-width dynamics are globally optimal, under some assumptions on the initialization of the weights. Our results establish optimality for feature-learning with wide RNNs in the mean-field regime
Provably efficient learning with typed parametric models
No full textTo quickly achieve good performance, reinforcement-learning algorithms for acting in large continuous-valued domains must use a representation that is both sufficiently powerful to capture important domain characteristics, and yet simultaneously allows generalization, or sharing, among experiences. Our algorithm balances this tradeoff by using a stochastic, switching, parametric dynamics representation. We argue that this model characterizes a number of significant, real-world domains, such as robot navigati on across varying terrain. We prove that this representational assumption allows our algorithm to be probably approximately correct with a sample complexity that scales polynomially with all problem-specific quantities including the state-space dimension. We also explicitly incorporate the error introduced by approximate planning in our sample complexity bounds, in contrast to prior Probably Approximately Correct (PAC) Markov Decision Processes (MDP) approaches, which typically assume the estimated MDP can be solved exactly. Our experimental results on constructing plans for driving to work using real car trajectory data, as well as a small robot experiment on navigating varying terrain, demonstrate that our dynamics representation enables us to capture real-world dynamics in a sufficient manner to produce good performance
XTab: Cross-table Pretraining for Tabular Transformers
No full textThe success of self-supervised learning in computer vision and natural language processing has motivated pretraining methods on tabular data. However, most existing tabular self-supervised learning models fail to leverage information across multiple data tables and cannot generalize to new tables. In this work, we introduce XTab, a framework for cross-table pretraining of tabular transformers on datasets from various domains. We address the challenge of inconsistent column types and quantities among tables by utilizing independent featurizers and using federated learning to pretrain the shared component. Tested on 84 tabular prediction tasks from the OpenML-AutoML Benchmark (AMLB), we show that (1) XTab consistently boosts the generalizability, learning speed, and performance of multiple tabular transformers, (2) by pretraining FT-Transformer via XTab, we achieve superior performance than other state-of-the-art tabular deep learning models on various tasks such as regression, binary, and multiclass classification.IN
Metagenomic Binning using Connectivity-constrained Variational Autoencoders
No full textCurrent state-of-the-art techniques for metage- nomic binning only utilize local features for the individual DNA sequences (contigs), neglecting additional information such as the assembly graph, in which the contigs are connected according to overlapping reads, and gene markers identified in the contigs. In this paper, we propose the use of a Variational AutoEncoder (VAE) tailored to leverage auxiliary structural information about contig relations when learning contig representations for subsequent metagenomic binning. Our method, CCVAE, improves on previous work that used VAEs for learning latent representations of the individual contigs, by constraining these representations according to the connectivity information from the assembly graph. Additionally, we incor- porate into the model additional information in the form of marker genes to better differentiate contigs from different genomes. Our experiments on both simulated and real-world datasets demon- strate that CCVAE outperforms current state-of- the-art techniques, thus providing a more effective method for metagenomic binning
- …
