arXiv subjects

Zachary C. Lipton

Publications and source records attributed to Zachary C. Lipton.

At least 127 records · Page 7Linked to original sources

arXiv2019

Embryo staging with weakly-supervised region selection and dynamically-decoded predictions

To optimize clinical outcomes, fertility clinics must strategically select which embryos to transfer. Common selection heuristics are formulas expressed in terms of the durations required to reach various developmental milestones, quantities historically annotated manually by experienced embryologists based on time-lapse EmbryoScope videos. We propose a new method for automatic embryo staging that exploits several sources of structure in this time-lapse data. First, noting that in each image the embryo occupies a small subregion, we jointly train a region proposal network with the downstream classifier to isolate the embryo. Notably, because we lack ground-truth bounding boxes, our we weakly supervise the region proposal network optimizing its parameters via reinforcement learning to improve the downstream classifier's loss. Moreover, noting that embryos reaching the blastocyst stage progress monotonically through earlier stages, we develop a dynamic-programming-based decoder that post-processes our predictions to select the most likely monotonic sequence of developmental stages. Our methods outperform vanilla residual networks and rival the best numbers in contemporary papers, as measured by both per-frame accuracy and transition prediction error, despite operating on smaller data than many.

cs.CV↗

arXiv2019

Learning Robust Representations by Projecting Superficial Statistics Out

Despite impressive performance as evaluated on i.i.d. holdout data, deep neural networks depend heavily on superficial statistics of the training data and are liable to break under distribution shift. For example, subtle changes to the background or texture of an image can break a seemingly powerful classifier. Building on previous work on domain generalization, we hope to produce a classifier that will generalize to previously unseen domains, even when domain identifiers are not available during training. This setting is challenging because the model may extract many distribution-specific (superficial) signals together with distribution-agnostic (semantic) signals. To overcome this challenge, we incorporate the gray-level co-occurrence matrix (GLCM) to extract patterns that our prior knowledge suggests are superficial: they are sensitive to the texture but unable to capture the gestalt of an image. Then we introduce two techniques for improving our networks' out-of-sample performance. The first method is built on the reverse gradient method that pushes our model to learn representations from which the GLCM representation is not predictable. The second method is built on the independence introduced by projecting the model's representation onto the subspace orthogonal to GLCM representation's. We test our method on the battery of standard domain generalization data sets and, interestingly, achieve comparable or better performance as compared to other domain generalization methods that explicitly require samples from the target distribution for training.

cs.CV↗

arXiv2019

Does mitigating ML's impact disparity require treatment disparity?

Following related work in law and policy, two notions of disparity have come to shape the study of fairness in algorithmic decision-making. Algorithms exhibit treatment disparity if they formally treat members of protected subgroups differently; algorithms exhibit impact disparity when outcomes differ across subgroups, even if the correlation arises unintentionally. Naturally, we can achieve impact parity through purposeful treatment disparity. In one thread of technical work, papers aim to reconcile the two forms of parity proposing disparate learning processes (DLPs). Here, the learning algorithm can see group membership during training but produce a classifier that is group-blind at test time. In this paper, we show theoretically that: (i) When other features correlate to group membership, DLPs will (indirectly) implement treatment disparity, undermining the policy desiderata they are designed to address; (ii) When group membership is partly revealed by other features, DLPs induce within-class discrimination; and (iii) In general, DLPs provide a suboptimal trade-off between accuracy and impact parity. Based on our technical analysis, we argue that transparent treatment disparity is preferable to occluded methods for achieving impact parity. Experimental results on several real-world datasets highlight the practical consequences of applying DLPs vs. per-group thresholds.

stat.ML↗

arXiv2018

Deep Bayesian Active Learning for Natural Language Processing: Results of a Large-Scale Empirical Study

Several recent papers investigate Active Learning (AL) for mitigating the data dependence of deep learning for natural language processing. However, the applicability of AL to real-world problems remains an open question. While in supervised learning, practitioners can try many different methods, evaluating each against a validation set before selecting a model, AL affords no such luxury. Over the course of one AL run, an agent annotates its dataset exhausting its labeling budget. Thus, given a new task, an active learner has no opportunity to compare models and acquisition functions. This paper provides a large scale empirical study of deep active learning, addressing multiple tasks and, for each, multiple datasets, multiple models, and a full suite of acquisition functions. We find that across all settings, Bayesian active learning by disagreement, using uncertainty estimates provided either by Dropout or Bayes-by Backprop significantly improves over i.i.d. baselines and usually outperforms classic uncertainty sampling.

cs.CL↗

arXiv2018

How Much Reading Does Reading Comprehension Require? A Critical Investigation of Popular Benchmarks

Many recent papers address reading comprehension, where examples consist of (question, passage, answer) tuples. Presumably, a model must combine information from both questions and passages to predict corresponding answers. However, despite intense interest in the topic, with hundreds of published papers vying for leaderboard dominance, basic questions about the difficulty of many popular benchmarks remain unanswered. In this paper, we establish sensible baselines for the bAbI, SQuAD, CBT, CNN, and Who-did-What datasets, finding that question- and passage-only models often perform surprisingly well. On $14$ out of $20$ bAbI tasks, passage-only models achieve greater than $50\%$ accuracy, sometimes matching the full model. Interestingly, while CBT provides $20$-sentence stories only the last is needed for comparably accurate prediction. By comparison, SQuAD and CNN appear better-constructed.

cs.CL↗

arXiv2018

Detecting and Correcting for Label Shift with Black Box Predictors

Faced with distribution shift between training and test set, we wish to detect and quantify the shift, and to correct our classifiers without test set labels. Motivated by medical diagnosis, where diseases (targets) cause symptoms (observations), we focus on label shift, where the label marginal $p(y)$ changes but the conditional $p(x| y)$ does not. We propose Black Box Shift Estimation (BBSE) to estimate the test distribution $p(y)$. BBSE exploits arbitrary black box predictors to reduce dimensionality prior to shift correction. While better predictors give tighter estimates, BBSE works even when predictors are biased, inaccurate, or uncalibrated, so long as their confusion matrices are invertible. We prove BBSE's consistency, bound its error, and introduce a statistical test that uses BBSE to detect shift. We also leverage BBSE to correct classifiers. Experiments demonstrate accurate estimates and improved prediction, even on high-dimensional datasets of natural images.

cs.LG↗

arXiv2018

Troubling Trends in Machine Learning Scholarship

Collectively, machine learning (ML) researchers are engaged in the creation and dissemination of knowledge about data-driven algorithms. In a given paper, researchers might aspire to any subset of the following goals, among others: to theoretically characterize what is learnable, to obtain understanding through empirically rigorous experiments, or to build a working system that has high predictive accuracy. While determining which knowledge warrants inquiry may be subjective, once the topic is fixed, papers are most valuable to the community when they act in service of the reader, creating foundational knowledge and communicating as clearly as possible. Recent progress in machine learning comes despite frequent departures from these ideals. In this paper, we focus on the following four patterns that appear to us to be trending in ML scholarship: (i) failure to distinguish between explanation and speculation; (ii) failure to identify the sources of empirical gains, e.g., emphasizing unnecessary modifications to neural architectures when gains actually stem from hyper-parameter tuning; (iii) mathiness: the use of mathematics that obfuscates or impresses rather than clarifies, e.g., by confusing technical and non-technical concepts; and (iv) misuse of language, e.g., by choosing terms of art with colloquial connotations or by overloading established technical terms. While the causes behind these patterns are uncertain, possibilities include the rapid expansion of the community, the consequent thinness of the reviewer pool, and the often-misaligned incentives between scholarship and short-term measures of success (e.g., bibliometrics, attention, and entrepreneurial opportunity). While each pattern offers a corresponding remedy (don't do it), we also discuss some speculative suggestions for how the community might combat these trends.

stat.ML↗

arXiv2018

Learning Noise-Invariant Representations for Robust Speech Recognition

Despite rapid advances in speech recognition, current models remain brittle to superficial perturbations to their inputs. Small amounts of noise can destroy the performance of an otherwise state-of-the-art model. To harden models against background noise, practitioners often perform data augmentation, adding artificially-noised examples to the training set, carrying over the original label. In this paper, we hypothesize that a clean example and its superficially perturbed counterparts shouldn't merely map to the same class --- they should map to the same representation. We propose invariant-representation-learning (IRL): At each training iteration, for each training example,we sample a noisy counterpart. We then apply a penalty term to coerce matched representations at each layer (above some chosen layer). Our key results, demonstrated on the Librispeech dataset are the following: (i) IRL significantly reduces character error rates (CER) on both 'clean' (3.3% vs 6.5%) and 'other' (11.0% vs 18.1%) test sets; (ii) on several out-of-domain noise settings (different from those seen during training), IRL's benefits are even more pronounced. Careful ablations confirm that our results are not simply due to shrinking activations at the chosen layers.

eess.AS↗

arXiv2018

Born Again Neural Networks

Knowledge Distillation (KD) consists of transferring â€œknowledgeâ€ from one machine learning model (the teacher) to another (the student). Commonly, the teacher is a high-capacity model with formidable performance, while the student is more compact. By transferring knowledge, one hopes to benefit from the studentâ€™s compactness, without sacrificing too much performance. We study KD from a new perspective: rather than compressing models, we train students parameterized identically to their teachers. Surprisingly, these Born-Again Networks (BANs), outperform their teachers significantly, both on computer vision and language modeling tasks. Our experiments with BANs based on DenseNets demonstrate state-of-the-art performance on the CIFAR-10 (3.5%) and CIFAR-100 (15.5%) datasets, by validation error. Additional experiments explore two distillation objectives: (i) Confidence-Weighted by Teacher Max (CWTM) and (ii) Dark Knowledge with Permuted Predictions (DKPP). Both methods elucidate the essential components of KD, demonstrating the effect of the teacher outputs on both predicted and non-predicted classes.

stat.ML↗

arXiv2018

Learning From Noisy Singly-labeled Data

Supervised learning depends on annotated examples, which are taken to be the \emph{ground truth}. But these labels often come from noisy crowdsourcing platforms, like Amazon Mechanical Turk. Practitioners typically collect multiple labels per example and aggregate the results to mitigate noise (the classic crowdsourcing problem). Given a fixed annotation budget and unlimited unlabeled data, redundant annotation comes at the expense of fewer labeled examples. This raises two fundamental questions: (1) How can we best learn from noisy workers? (2) How should we allocate our labeling budget to maximize the performance of a classifier? We propose a new algorithm for jointly modeling labels and worker quality from noisy crowd-sourced data. The alternating minimization proceeds in rounds, estimating worker quality from disagreement with the current model and then updating the model by optimizing a loss function that accounts for the current estimate of worker quality. Unlike previous approaches, even with only one annotation per example, our algorithm can estimate worker quality. We establish a generalization error bound for models learned with our algorithm and establish theoretically that it's better to label many examples once (vs less multiply) when worker quality is above a threshold. Experiments conducted on both ImageNet (with simulated noisy workers) and MS-COCO (using the real crowdsourced labels) confirm our algorithm's benefits.

cs.LG↗

arXiv2018

Improving Factor-Based Quantitative Investing by Forecasting Company Fundamentals

On a periodic basis, publicly traded companies are required to report fundamentals: financial data such as revenue, operating income, debt, among others. These data points provide some insight into the financial health of a company. Academic research has identified some factors, i.e. computed features of the reported data, that are known through retrospective analysis to outperform the market average. Two popular factors are the book value normalized by market capitalization (book-to-market) and the operating income normalized by the enterprise value (EBIT/EV). In this paper: we first show through simulation that if we could (clairvoyantly) select stocks using factors calculated on future fundamentals (via oracle), then our portfolios would far outperform a standard factor approach. Motivated by this analysis, we train deep neural networks to forecast future fundamentals based on a trailing 5-years window. Quantitative analysis demonstrates a significant improvement in MSE over a naive strategy. Moreover, in retrospective analysis using an industry-grade stock portfolio simulator (backtester), we show an improvement in compounded annual return to 17.1% (MLP) vs 14.4% for a standard factor model.

stat.ML↗

arXiv2018

Combating Reinforcement Learning's Sisyphean Curse with Intrinsic Fear

Many practical environments contain catastrophic states that an optimal agent would visit infrequently or never. Even on toy problems, Deep Reinforcement Learning (DRL) agents tend to periodically revisit these states upon forgetting their existence under a new policy. We introduce intrinsic fear (IF), a learned reward shaping that guards DRL agents against periodic catastrophes. IF agents possess a fear model trained to predict the probability of imminent catastrophe. This score is then used to penalize the Q-learning objective. Our theoretical analysis bounds the reduction in average return due to learning on the perturbed objective. We also prove robustness to classification errors. As a bonus, IF models tend to learn faster, owing to reward shaping. Experiments demonstrate that intrinsic-fear DQNs solve otherwise pathological environments and improve on several Atari games.

cs.LG↗

arXiv2018

Correction by Projection: Denoising Images with Generative Adversarial Networks

Generative adversarial networks (GANs) transform low-dimensional latent vectors into visually plausible images. If the real dataset contains only clean images, then ostensibly, the manifold learned by the GAN should contain only clean images. In this paper, we propose to denoise corrupted images by finding the nearest point on the GAN manifold, recovering latent vectors by minimizing distances in image space. We first demonstrate that given a corrupted version of an image that truly lies on the GAN manifold, we can approximately recover the latent vector and denoise the image, obtaining significantly higher quality, comparing with BM3D. Next, we demonstrate that latent vectors recovered from noisy images exhibit a consistent bias. By subtracting this bias before projecting back to image space, we improve denoising results even further. Finally, even for unseen images, our method performs better at denoising better than BM3D. Notably, the basic version of our method (without bias correction) requires no prior knowledge on the noise variance. To achieve the highest possible denoising quality, the best performing signal processing based methods, such as BM3D, require an estimate of the blur kernel.

cs.CV↗

arXiv2018

Stochastic Activation Pruning for Robust Adversarial Defense

Neural networks are known to be vulnerable to adversarial examples. Carefully chosen perturbations to real images, while imperceptible to humans, induce misclassification and threaten the reliability of deep learning systems in the wild. To guard against adversarial examples, we take inspiration from game theory and cast the problem as a minimax zero-sum game between the adversary and the model. In general, for such games, the optimal strategy for both players requires a stochastic policy, also known as a mixed strategy. In this light, we propose Stochastic Activation Pruning (SAP), a mixed strategy for adversarial defense. SAP prunes a random subset of activations (preferentially pruning those with smaller magnitude) and scales up the survivors to compensate. We can apply SAP to pretrained networks, including adversarially trained models, without fine-tuning, providing robustness against adversarial examples. Experiments demonstrate that SAP confers robustness against attacks, increasing accuracy and preserving calibration.

cs.LG↗

arXiv2018

Semantically Decomposing the Latent Spaces of Generative Adversarial Networks

We propose a new algorithm for training generative adversarial networks that jointly learns latent codes for both identities (e.g. individual humans) and observations (e.g. specific photographs). By fixing the identity portion of the latent codes, we can generate diverse images of the same subject, and by fixing the observation portion, we can traverse the manifold of subjects while maintaining contingent aspects such as lighting and pose. Our algorithm features a pairwise training scheme in which each sample from the generator consists of two images with a common identity code. Corresponding samples from the real dataset consist of two distinct photographs of the same subject. In order to fool the discriminator, the generator must produce pairs that are photorealistic, distinct, and appear to depict the same individual. We augment both the DCGAN and BEGAN approaches with Siamese discriminators to facilitate pairwise training. Experiments with human judges and an off-the-shelf face verification system demonstrate our algorithm's ability to generate convincing, identity-matched photographs.

cs.LG↗

arXiv2018

Deep Active Learning for Named Entity Recognition

Deep learning has yielded state-of-the-art performance on many natural language processing tasks including named entity recognition (NER). However, this typically requires large amounts of labeled data. In this work, we demonstrate that the amount of labeled training data can be drastically reduced when deep learning is combined with active learning. While active learning is sample-efficient, it can be computationally expensive since it requires iterative retraining. To speed this up, we introduce a lightweight architecture for NER, viz., the CNN-CNN-LSTM model consisting of convolutional character and word encoders and a long short term memory (LSTM) tag decoder. The model achieves nearly state-of-the-art performance on standard datasets for the task while being computationally much more efficient than best performing models. We carry out incremental active learning, during the training process, and are able to nearly match state-of-the-art performance with just 25\% of the original training data.

cs.CL↗

arXiv2017

The Doctor Just Won't Accept That!

Calls to arms to build interpretable models express a well-founded discomfort with machine learning. Should a software agent that does not even know what a loan is decide who qualifies for one? Indeed, we ought to be cautious about injecting machine learning (or anything else, for that matter) into applications where there may be a significant risk of causing social harm. However, claims that stakeholders "just won't accept that!" do not provide a sufficient foundation for a proposed field of study. For the field of interpretable machine learning to advance, we must ask the following questions: What precisely won't various stakeholders accept? What do they want? Are these desiderata reasonable? Are they feasible? In order to answer these questions, we'll have to give real-world problems and their respective stakeholders greater consideration.

stat.ML↗

arXiv2017

BBQ-Networks: Efficient Exploration in Deep Reinforcement Learning for Task-Oriented Dialogue Systems

We present a new algorithm that significantly improves the efficiency of exploration for deep Q-learning agents in dialogue systems. Our agents explore via Thompson sampling, drawing Monte Carlo samples from a Bayes-by-Backprop neural network. Our algorithm learns much faster than common exploration strategies such as $ε$-greedy, Boltzmann, bootstrapping, and intrinsic-reward-based ones. Additionally, we show that spiking the replay buffer with experiences from just a few successful episodes can make Q-learning feasible when it might otherwise fail.

cs.LG↗