arXiv subjects

Roy Schwartz

Publications and source records attributed to Roy Schwartz.

At least 73 records · Page 4Linked to original sources

arXiv2021

Automatic Generation of Contrast Sets from Scene Graphs: Probing the Compositional Consistency of GQA

Recent works have shown that supervised models often exploit data artifacts to achieve good test scores while their performance severely degrades on samples outside their training distribution. Contrast sets (Gardneret al., 2020) quantify this phenomenon by perturbing test samples in a minimal way such that the output label is modified. While most contrast sets were created manually, requiring intensive annotation effort, we present a novel method which leverages rich semantic input representation to automatically generate contrast sets for the visual question answering task. Our method computes the answer of perturbed questions, thus vastly reducing annotation cost and enabling thorough evaluation of models' performance on various semantic aspects (e.g., spatial or relational reasoning). We demonstrate the effectiveness of our approach on the GQA dataset and its semantic scene graph image representation. We find that, despite GQA's compositionality and carefully balanced label distribution, two high-performing models drop 13-17% in accuracy compared to the original test set. Finally, we show that our automatic perturbation can be applied to the training set to mitigate the degradation in performance, opening the door to more robust models.

cs.CL↗

arXiv2021

A Refined Analysis of Submodular Greedy

Many algorithms for maximizing a monotone submodular function subject to a knapsack constraint rely on the natural greedy heuristic. We present a novel refined analysis of this greedy heuristic which enables us to: $(1)$ reduce the enumeration in the tight $(1-e^{-1})$-approximation of [Sviridenko 04] from subsets of size three to two; $(2)$ present an improved upper bound of $0.42945$ for the classic algorithm which returns the better between a single element and the output of the greedy heuristic.

cs.DS↗

arXiv2020

Dataset Cartography: Mapping and Diagnosing Datasets with Training Dynamics

Large datasets have become commonplace in NLP research. However, the increased emphasis on data quantity has made it challenging to assess the quality of data. We introduce Data Maps---a model-based tool to characterize and diagnose datasets. We leverage a largely ignored source of information: the behavior of the model on individual instances during training (training dynamics) for building data maps. This yields two intuitive measures for each example---the model's confidence in the true class, and the variability of this confidence across epochs---obtained in a single run of training. Experiments across four datasets show that these model-dependent measures reveal three distinct regions in the data map, each with pronounced characteristics. First, our data maps show the presence of "ambiguous" regions with respect to the model, which contribute the most towards out-of-distribution generalization. Second, the most populous regions in the data are "easy to learn" for the model, and play an important role in model optimization. Finally, data maps uncover a region with instances that the model finds "hard to learn"; these often correspond to labeling errors. Our results indicate that a shift in focus from quantity to quality of data could lead to robust models and improved out-of-distribution generalization.

cs.CL↗

arXiv2020

A Formal Hierarchy of RNN Architectures

We develop a formal hierarchy of the expressive capacity of RNN architectures. The hierarchy is based on two formal properties: space complexity, which measures the RNN's memory, and rational recurrence, defined as whether the recurrent update can be described by a weighted finite-state machine. We place several RNN variants within this hierarchy. For example, we prove the LSTM is not rational, which formally separates it from the related QRNN (Bradbury et al., 2016). We also show how these models' expressive capacity is expanded by stacking multiple layers or composing them with different pooling functions. Our results build on the theory of "saturated" RNNs (Merrill, 2019). While formally extending these findings to unsaturated RNNs is left to future work, we hypothesize that the practical learnable capacity of unsaturated RNNs obeys a similar hierarchy. Experimental findings from training unsaturated networks on formal languages support this conjecture.

cs.CL↗

arXiv2020

A Mixture of $h-1$ Heads is Better than $h$ Heads

Multi-head attentive neural architectures have achieved state-of-the-art results on a variety of natural language processing tasks. Evidence has shown that they are overparameterized; attention heads can be pruned without significant performance loss. In this work, we instead "reallocate" them -- the model learns to activate different heads on different inputs. Drawing connections between multi-head attention and mixture of experts, we propose the mixture of attentive experts model (MAE). MAE is trained using a block coordinate descent algorithm that alternates between updating (1) the responsibilities of the experts and (2) their parameters. Experiments on machine translation and language modeling show that MAE outperforms strong baselines on both tasks. Particularly, on the WMT14 English to German translation dataset, MAE improves over "transformer-base" by 0.8 BLEU, with a comparable number of parameters. Our analysis shows that our model learns to specialize different experts to different inputs.

cs.CL↗

arXiv2020

The Right Tool for the Job: Matching Model and Instance Complexities

As NLP models become larger, executing a trained model requires significant computational resources incurring monetary and environmental costs. To better respect a given inference budget, we propose a modification to contextual representation fine-tuning which, during inference, allows for an early (and fast) "exit" from neural network calculations for simple instances, and late (and accurate) exit for hard instances. To achieve this, we add classifiers to different layers of BERT and use their calibrated confidence scores to make early exit decisions. We test our proposed modification on five different datasets in two tasks: three text classification datasets and two natural language inference benchmarks. Our method presents a favorable speed/accuracy tradeoff in almost all cases, producing models which are up to five times faster than the state of the art, while preserving their accuracy. Our method also requires almost no additional training resources (in either time or parameters) compared to the baseline BERT model. Finally, our method alleviates the need for costly retraining of multiple models at different levels of efficiency; we allow users to control the inference speed/accuracy tradeoff using a single trained model, by setting a single variable at inference time. We publicly release our code.

cs.CL↗

arXiv2020

Fine-Tuning Pretrained Language Models: Weight Initializations, Data Orders, and Early Stopping

Fine-tuning pretrained contextual word embedding models to supervised downstream tasks has become commonplace in natural language processing. This process, however, is often brittle: even with the same hyperparameter values, distinct random seeds can lead to substantially different results. To better understand this phenomenon, we experiment with four datasets from the GLUE benchmark, fine-tuning BERT hundreds of times on each while varying only the random seeds. We find substantial performance increases compared to previously reported results, and we quantify how the performance of the best-found model varies as a function of the number of fine-tuning trials. Further, we examine two factors influenced by the choice of random seed: weight initialization and training data order. We find that both contribute comparably to the variance of out-of-sample performance, and that some weight initializations perform well across all tasks explored. On small datasets, we observe that many fine-tuning trials diverge part of the way through training, and we offer best practices for practitioners to stop training less promising runs early. We publicly release all of our experimental data, including training and validation scores for 2,100 trials, to encourage further analysis of training dynamics during fine-tuning.

cs.CL↗

arXiv2020

Fair Correlation Clustering

In this paper we study the problem of correlation clustering under fairness constraints. In the classic correlation clustering problem, we are given a complete graph where each edge is labeled positive or negative. The goal is to obtain a clustering of the vertices that minimizes disagreements -- the number of negative edges trapped inside a cluster plus positive edges between different clusters. We consider two variations of fairness constraint for the problem of correlation clustering where each node has a color, and the goal is to form clusters that do not over-represent vertices of any color. The first variant aims to generate clusters with minimum disagreements, where the distribution of a feature (e.g. gender) in each cluster is same as the global distribution. For the case of two colors when the desired ratio of the number of colors in each cluster is $1:p$, we get $\mathcal{O}(p^2)$-approximation algorithm. Our algorithm could be extended to the case of multiple colors. We prove this problem is NP-hard. The second variant considers relative upper and lower bounds on the number of nodes of any color in a cluster. The goal is to avoid violating upper and lower bounds corresponding to each color in each cluster while minimizing the total number of disagreements. Along with our theoretical results, we show the effectiveness of our algorithm to generate fair clusters by empirical evaluation on real world data sets.

cs.DS↗

arXiv2019

Knowledge Enhanced Contextual Word Representations

Contextual word representations, typically trained on unstructured, unlabeled text, do not contain any explicit grounding to real world entities and are often unable to remember facts about those entities. We propose a general method to embed multiple knowledge bases (KBs) into large scale models, and thereby enhance their representations with structured, human-curated knowledge. For each KB, we first use an integrated entity linker to retrieve relevant entity embeddings, then update contextual word representations via a form of word-to-entity attention. In contrast to previous approaches, the entity linkers and self-supervised language modeling objective are jointly trained end-to-end in a multitask setting that combines a small amount of entity linking supervision with a large amount of raw text. After integrating WordNet and a subset of Wikipedia into BERT, the knowledge enhanced BERT (KnowBert) demonstrates improved perplexity, ability to recall facts as measured in a probing task and downstream performance on relationship extraction, entity typing, and word sense disambiguation. KnowBert's runtime is comparable to BERT's and it scales to large KBs.

cs.CL↗

arXiv2019

Sticky Brownian Rounding and its Applications to Constraint Satisfaction Problems

Semidefinite programming is a powerful tool in the design and analysis of approximation algorithms for combinatorial optimization problems. In particular, the random hyperplane rounding method of Goemans and Williamson has been extensively studied for more than two decades, resulting in various extensions to the original technique and beautiful algorithms for a wide range of applications. Despite the fact that this approach yields tight approximation guarantees for some problems, e.g., Max-Cut, for many others, e.g., Max-SAT and Max-DiCut, the tight approximation ratio is still unknown. One of the main reasons for this is the fact that very few techniques for rounding semidefinite relaxations are known. In this work, we present a new general and simple method for rounding semi-definite programs, based on Brownian motion. Our approach is inspired by recent results in algorithmic discrepancy theory. We develop and present tools for analyzing our new rounding algorithms, utilizing mathematical machinery from the theory of Brownian motion, complex analysis, and partial differential equations. Focusing on constraint satisfaction problems, we apply our method to several classical problems, including Max-Cut, Max-2SAT, and MaxDiCut, and derive new algorithms that are competitive with the best known results. To illustrate the versatility and general applicability of our approach, we give new approximation algorithms for the Max-Cut problem with side constraints that crucially utilizes measure concentration results for the Sticky Brownian Motion, a feature missing from hyperplane rounding and its generalizations

cs.DS↗

arXiv2019

Show Your Work: Improved Reporting of Experimental Results

Research in natural language processing proceeds, in part, by demonstrating that new models achieve superior performance (e.g., accuracy) on held-out test data, compared to previous results. In this paper, we demonstrate that test-set performance scores alone are insufficient for drawing accurate conclusions about which model performs best. We argue for reporting additional details, especially performance on validation data obtained during model development. We present a novel technique for doing so: expected validation performance of the best-found model as a function of computation budget (i.e., the number of hyperparameter search trials or the overall training time). Using our approach, we find multiple recent model comparisons where authors would have reached a different conclusion if they had used more (or less) computation. Our approach also allows us to estimate the amount of computation required to obtain a given accuracy; applying it to several recently published results yields massive variation across papers, from hours to weeks. We conclude with a set of best practices for reporting experimental results which allow for robust future comparisons, and provide code to allow researchers to use our technique.

cs.LG↗

arXiv2019

RNN Architecture Learning with Sparse Regularization

Neural models for NLP typically use large numbers of parameters to reach state-of-the-art performance, which can lead to excessive memory usage and increased runtime. We present a structure learning method for learning sparse, parameter-efficient NLP models. Our method applies group lasso to rational RNNs (Peng et al., 2018), a family of models that is closely connected to weighted finite-state automata (WFSAs). We take advantage of rational RNNs' natural grouping of the weights, so the group lasso penalty directly removes WFSA states, substantially reducing the number of parameters in the model. Our experiments on a number of sentiment analysis datasets, using both GloVe and BERT embeddings, show that our approach learns neural structures which have fewer parameters without sacrificing performance relative to parameter-rich baselines. Our method also highlights the interpretable properties of rational RNNs. We show that sparsifying such models makes them easier to visualize, and we present models that rely exclusively on as few as three WFSAs after pruning more than 90% of the weights. We publicly release our code.

cs.CL↗

arXiv2019

PaLM: A Hybrid Parser and Language Model

We present PaLM, a hybrid parser and neural language model. Building on an RNN language model, PaLM adds an attention layer over text spans in the left context. An unsupervised constituency parser can be derived from its attention weights, using a greedy decoding algorithm. We evaluate PaLM on language modeling, and empirically show that it outperforms strong baselines. If syntactic annotations are available, the attention component can be trained in a supervised manner, providing syntactically-informed representations of the context, and further improving language modeling performance.

cs.CL↗

arXiv2019

Green AI

The computations required for deep learning research have been doubling every few months, resulting in an estimated 300,000x increase from 2012 to 2018 [2]. These computations have a surprisingly large carbon footprint [38]. Ironically, deep learning was inspired by the human brain, which is remarkably energy efficient. Moreover, the financial cost of the computations can make it difficult for academics, students, and researchers, in particular those from emerging economies, to engage in deep learning research. This position paper advocates a practical solution by making efficiency an evaluation criterion for research alongside accuracy and related measures. In addition, we propose reporting the financial cost or "price tag" of developing, training, and running models to provide baselines for the investigation of increasingly efficient methods. Our goal is to make AI both greener and more inclusive---enabling any inspired undergraduate with a laptop to write high-quality research papers. Green AI is an emerging focus at the Allen Institute for AI.

cs.CY↗

arXiv2019

Min-Max Correlation Clustering via MultiCut

Correlation clustering is a fundamental combinatorial optimization problem arising in many contexts and applications that has been the subject of dozens of papers in the literature. In this problem we are given a general weighted graph where each edge is labeled positive or negative. The goal is to obtain a partitioning (clustering) of the vertices that minimizes disagreements - weight of negative edges trapped inside a cluster plus positive edges between different clusters. Most of the papers on this topic mainly focus on minimizing total disagreement, a global objective for this problem. In this paper, we study a cluster-wise objective function that asks to minimize the maximum number of disagreements of each cluster, which we call min-max correlation clustering. The min-max objective is a natural objective that respects the quality of every cluster. In this paper, we provide the first nontrivial approximation algorithm for this problem achieving an $\mathcal{O}(\sqrt{\log n\cdot\max\{\log(|E^-|),\log(k)\}})$ approximation for general weighted graphs, where $|E^-|$ denotes the number of negative edges and $k$ is the number of clusters in the optimum solution. To do so, we also obtain a corresponding result for multicut where we wish to find a multicut solution while trying to minimize the total weight of cut edges on every component. The results are then further improved to obtain (i) $\mathcal{O}(r^2)$-approximation for min-max correlation clustering and min-max multicut for graphs that exclude $K_{r,r}$ minors (ii) a 14-approximation for the min-max correlation clustering on complete graphs.

cs.DS↗

arXiv2019

Online and Offline Greedy Algorithms for Routing with Switching Costs

Motivated by the use of high speed circuit switches in large scale data centers, we consider the problem of circuit switch scheduling. In this problem we are given demands between pairs of servers and the goal is to schedule at every time step a matching between the servers while maximizing the total satisfied demand over time. The crux of this scheduling problem is that once one shifts from one matching to a different one a fixed delay $δ$ is incurred during which no data can be transmitted. For the offline version of the problem we present a $(1-\frac{1}{e}-ε)$ approximation ratio (for any constant $ε>0$). Since the natural linear programming relaxation for the problem has an unbounded integrality gap, we adopt a hybrid approach that combines the combinatorial greedy with randomized rounding of a different suitable linear program. For the online version of the problem we present a (bi-criteria) $ ((e-1)/(2e-1)-ε)$-competitive ratio (for any constant $ε>0$ ) that exceeds time by an additive factor of $O(\fracδε)$. We note that no uni-criteria online algorithm is possible. Surprisingly, we obtain the result by reducing the online version to the offline one.

cs.DS↗

arXiv2019

Inoculation by Fine-Tuning: A Method for Analyzing Challenge Datasets

Several datasets have recently been constructed to expose brittleness in models trained on existing benchmarks. While model performance on these challenge datasets is significantly lower compared to the original benchmark, it is unclear what particular weaknesses they reveal. For example, a challenge dataset may be difficult because it targets phenomena that current models cannot capture, or because it simply exploits blind spots in a model's specific training set. We introduce inoculation by fine-tuning, a new analysis method for studying challenge datasets by exposing models (the metaphorical patient) to a small amount of data from the challenge dataset (a metaphorical pathogen) and assessing how well they can adapt. We apply our method to analyze the NLI "stress tests" (Naik et al., 2018) and the Adversarial SQuAD dataset (Jia and Liang, 2017). We show that after slight exposure, some of these datasets are no longer challenging, while others remain difficult. Our results indicate that failures on challenge datasets may lead to very different conclusions about models, training datasets, and the challenge datasets themselves.

cs.CL↗

arXiv2019

LSTMs Exploit Linguistic Attributes of Data

While recurrent neural networks have found success in a variety of natural language processing applications, they are general models of sequential data. We investigate how the properties of natural language data affect an LSTM's ability to learn a nonlinguistic task: recalling elements from its input. We find that models trained on natural language data are able to recall tokens from much longer sequences than models trained on non-language sequential data. Furthermore, we show that the LSTM learns to solve the memorization task by explicitly using a subset of its neurons to count timesteps in the input. We hypothesize that the patterns and structure in natural language data enable LSTMs to learn by providing approximate ways of reducing loss, but understanding the effect of different training data on the learnability of LSTMs remains an open question.

cs.CL↗