DOCSLIB.ORG

Weak Supervision from High-Level Abstractions A

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Weak Supervision from High-Level Abstractions A WEAK SUPERVISION FROM HIGH-LEVEL ABSTRACTIONS A DISSERTATION SUBMITTED TO THE DEPARTMENT OF COMPUTER SCIENCE AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY Braden Jay Hancock August 2019 © 2019 by Braden Jay Hancock. All Rights Reserved. Re-distributed by Stanford University under license with the author. This work is licensed under a Creative Commons Attribution- Noncommercial 3.0 United States License. http://creativecommons.org/licenses/by-nc/3.0/us/ This dissertation is online at: http://purl.stanford.edu/ns523jd4552 ii I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy. Chris Re, Primary Adviser I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy. Dan Jurafsky I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy. Percy Liang Approved for the Stanford University Committee on Graduate Studies. Patricia J. Gumport, Vice Provost for Graduate Education This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file in University Archives. iii Abstract The interfaces for interacting with machine learning models are changing. Consider, for example, that while computers run on 1s and 0s, that is no longer the level of abstraction we use to program most computers. Instead, we use higher-level abstractions such as assembly language, high-level languages, or declarative languages to more efficiently convert our objectives into code. Similarly, most machine learning models are trained with "1s and 0s" (individually labeled examples), but we need not limit ourselves to interacting with them at this low level. Instead, we can use higher-level abstractions to more efficiently convert our domain knowledge into the inputs our models require. In this work, we show that weak supervision from high-level abstractions can be used to train high-performance machine learning models. At three different levels of abstraction, we describe the system we built to enable such interaction. We begin with Snorkel, which elevates label generation from a manual process to a programmatic one. With this system, domain experts encode their knowledge in potentially noisy and correlated black-box functions called labeling functions. These functions can then be automatically denoised and applied to unlabeled data to create large training sets quickly. Next, with Fonduer we enable an abstraction one step higher where advanced primitives defined over multiple modalities (visual, textual, structural, and tabular) allow users to programmatically supervise over richly formatted data (e.g., PDFs with tables and formatting). Finally, in BabbleLabble we show that we can even utilize supervision given in the form of natural language explanations, maintaining the benefits of programmatic supervision while removing the burden of writing code. For all of these systems, we demonstrate their effectiveness with empirical results and present real-world use cases where they have enabled rapid development of machine learning applications, including in bio-medicine, commerce, and defense. iv Acknowledgments My Ph.D. experience was rich and rewarding, and for that I owe a great debt to a great many people. I am deeply grateful to my advisor, Chris Ré. In an advisor I hoped for a master in identifying problems where progress would translate into real-world impact. I absolutely found this in Chris, but much more. From him I internalized valuable lessons such as: focus on process, not products; a paper is a receipt for good work, not the good work itself; and the reward for hard work is always more hard work. No one works harder than Chris, and it was an honor to be a part of his lab during my tenure at Stanford. I am grateful to Percy Liang and Dan Jurafsky, whose classes and insights during my first year at Stanford gave me a love for natural language processing and appreciation for education at the highest level. I owe a great deal to my fellow students and labmates in the Hazy Research group. It takes a village to raise a researcher, and innumerable paper swaps, whiteboard discussions, and late night hackathons together have made me the researcher I am today. In particular, I would like to acknowledge my fellow Ph.D. students Alex Ratner and Paroma Varma, who were with me through it all. I have had outstanding mentors at every stage of my education who took chances on me when they certainly did not have to: John Clark at AFRL before I knew a thing about programming, Christopher Mattson at BYU before I knew a thing about research, Mark Dredze at Johns Hopkins before I knew a thing about NLP, Vijay Gadepally at MIT Lincoln Laboratory before I knew a thing about machine learning, Hongrae Lee at Google before I knew a thing about production environments, and Antoine Bordes and Jason Weston at Facebook before I knew a thing about dialogue. My research would not have been possible without the financial support of my funders: the NSF Graduate Research Fellowship and Stanford Finch Family Fellowship especially, but also the DOE, NIH, ONR, DARPA, member companies of Stanford DAWN, and many other organizations supporting the Hazy Research group. Finally and most of all, I am grateful for my wife Lauren and daughters Annie and Pippa. When we moved to Stanford with a 10-day-old baby, I had hope but no assurance that I would be able to keep up with the rigorous demands of a top-tier doctoral program while simultaneously learning how to raise a family. I could not have anticipated then how many times more beautiful and rich my experience would be because it was shared with them. v Contents Abstract iv Acknowledgments v 1 Introduction 1 2 Weak Supervision from Code 4 2.1 Introduction . .4 2.2 Snorkel Architecture . .8 2.2.1 A Language for Weak Supervision . 10 2.2.2 Generative Model . 14 2.2.3 Discriminative Model . 15 2.3 Weak Supervision Tradeoffs . 15 2.3.1 Modeling Accuracies . 16 2.3.2 Modeling Structure . 20 2.4 Evaluation . 23 2.4.1 Applications . 24 2.4.2 User Study . 31 2.5 Extensions & Next Steps . 34 2.5.1 Extensions for Real-World Deployments . 34 2.5.2 Multi-Task Weak Supervision . 34 2.5.3 Future Directions . 35 2.6 Related Work . 35 2.7 Conclusion . 36 3 Weak Supervision from Primitives 38 3.1 Introduction . 38 3.2 Background . 42 3.2.1 Knowledge Base Construction . 42 vi 3.2.2 Recurrent Neural Networks . 43 3.3 The Fonduer Framework . 45 3.3.1 Fonduer’s Data Model . 45 3.3.2 User Inputs and Fonduer’s Pipeline . 46 3.3.3 Fonduer’s Programming Model for KBC . 49 3.4 KBC in Fonduer ........................................ 50 3.4.1 Candidate Generation . 50 3.4.2 Multimodal LSTM Model . 51 3.4.3 Multimodal Supervision . 54 3.5 Experiments . 54 3.5.1 Experimental Settings . 54 3.5.2 Experimental Results . 56 3.5.3 Ablation Studies . 58 3.6 User Study . 61 3.7 Extensions . 63 3.8 Related Work . 64 3.9 Conclusion . 64 4 Weak Supervision from Natural Language 65 4.1 Introduction . 65 4.2 The BabbleLabble Framework . 67 4.2.1 Explanations . 68 4.2.2 Semantic Parser . 68 4.2.3 Filter Bank . 69 4.2.4 Label Aggregator . 70 4.2.5 Discriminative Model . 71 4.3 Experimental Setup . 71 4.3.1 Datasets . 72 4.3.2 Experimental Settings . 73 4.4 Experimental Results . 73 4.4.1 High Bandwidth Supervision . 73 4.4.2 Utility of Incorrect Parses . 74 4.4.3 Using LFs as Functions or Features . 75 4.5 Related Work and Discussion . 76 4.6 Extensions . 77 vii 5 Discussion & Conclusion 78 5.1 Advantages of Programmatic Supervision . 78 5.2 Limitations . 79 5.3 The Supervision Stack . 80 5.4 Conclusion . 81 A Snorkel Appendix 82 A.1 Additional Material for Sec. 3.1 . 82 A.1.1 Minor Notes . 82 A.1.2 Proof of Proposition 1 . 82 A.1.3 Proof of Proposition 2 . 84 A.1.4 Proof of Proposition 3 . 85 B Fonduer Appendix 88 B.1 Data Programming . 88 B.1.1 Components of Data Programming . 88 B.1.2 Theoretical Guarantees . 89 B.2 Extended Feature Library . ..
Load more

Recommended publications
  • Generating Multi-Agent Trajectories Using Programmatic Weak
    Published as a conference paper at ICLR 2019 GENERATING MULTI-AGENT TRAJECTORIES USING PROGRAMMATIC WEAK SUPERVISION Eric Zhan Stephan Zheng∗ Caltech Salesforce [email protected] [email protected] Yisong Yue Long Sha & Patrick Lucey Caltech STATS [email protected] flsha,[email protected] ABSTRACT We study the problem of training sequential generative models for capturing co- ordinated multi-agent trajectory behavior, such as offensive basketball gameplay. When modeling such settings, it is often beneficial to design hierarchical mod- els that can capture long-term coordination using intermediate variables. Further- more, these intermediate variables should capture interesting high-level behavioral semantics in an interpretable and manipulatable way. We present a hierarchical framework that can effectively learn such sequential generative models. Our ap- proach is inspired by recent work on leveraging programmatically produced weak labels, which we extend to the spatiotemporal regime. In addition to synthetic settings, we show how to instantiate our framework to effectively model complex interactions between basketball players and generate realistic multi-agent trajec- tories of basketball gameplay over long time periods. We validate our approach using both quantitative and qualitative evaluations, including a user study compar- ison conducted with professional sports analysts.1 1 INTRODUCTION The ongoing explosion of recorded tracking data is enabling the study of fine-grained behavior in many domains: sports (Miller et al., 2014; Yue et al., 2014; Zheng et al., 2016; Le et al., 2017), video games (Ross et al., 2011), video & motion capture (Suwajanakorn et al., 2017; Taylor et al., 2017; Xue et al., 2016), navigation & driving (Ziebart et al., 2009; Zhang & Cho, 2017; Li et al., 2017), lab- oratory animal behaviors (Johnson et al., 2016; Eyjolfsdottir et al., 2017), and tele-operated robotics (Abbeel & Ng, 2004; Lin et al., 2006).
    [Show full text]
  • Constrained Labeling for Weakly Supervised Learning
    Constrained Labeling for Weakly Supervised Learning Chidubem Arachie1 Bert Huang2 1Department of Computer Science, Virginia Tech, Blacksburg, Virginia, USA 2Department of Computer Science, Data Intensive Studies Center, Tufts University, Medford, Massachusetts, USA Abstract sion. Weakly supervised learning involves training models using noisy labels. Using multiple sources or forms of weak supervision is common, as it provides diverse information Curation of large fully supervised datasets has be- to the model. However, each source of weak supervision has come one of the major roadblocks for machine its own bias that can be transmitted to the model. Different learning. Weak supervision provides an alterna- weak supervision signals can also conflict, overlap, or—in tive to supervised learning by training with cheap, the worst case—make dependent errors. Thus, a naive com- noisy, and possibly correlated labeling functions bination of these weak signals would hurt the quality of from varying sources. The key challenge in weakly a learned model. The key problem then is how to reliably supervised learning is combining the different combine various sources of weak signals to train an accurate weak supervision signals while navigating mislead- model. ing correlations in their errors. In this paper, we propose a simple data-free approach for combin- To solve this problem, we propose constrained label learn- ing weak supervision signals by defining a con- ing (CLL), a method that processes various weak supervi- strained space for the possible labels of the weak sion signals and combines them to produce high-quality signals and training with a random labeling within training labels. The idea behind CLL is that, given the weak this constrained space.
    [Show full text]
  • Multi-Task Weak Supervision Enables Automated Abnormality Localization in Whole-Body FDG-PET/CT
    Multi-task weak supervision enables automated abnormality localization in whole-body FDG-PET/CT Sabri Eyuboglu∗1, Geoffrey Angus∗1, Bhavik N. Patel2, Anuj Pareek2, Guido Davidzon2, Jared Dunnmon,∗∗1 Matthew P. Lungren ∗∗2 1 Department of Computer Science, Stanford University, Stanford, CA 94305, USA 2 Department of Radiology, Stanford University, Stanford, CA 94305, USA ∗ ∗∗ Equal contribution; Computational decision support systems could provide clinical value in whole-body FDG- PET/CT workflows. However, the lack of labeled data combined with the sheer size of each scan make it challenging to apply existing supervised machine learning systems. Leveraging recent advancements in natural language processing, we develop a weak supervision frame- work that extracts imperfect, yet highly granular regional abnormality labels from the radi- ologist reports associated with each scan. Our framework automatically labels each region in a custom ontology of anatomical regions, providing a structured profile of the pathologies in each scan. We design an attention-based, multi-task CNN architecture that can be trained using these generated labels to localize abnormalities in whole-body scans. We show that our multi-task representation is critical for strong performance on rare abnormalities for which we have little data and for accurately predicting mortality from imaging data alone (AUROC 80%), a critical challenge for palliative care teams. 1 Introduction The availability of large, labeled datasets has fueled recent progress in machine learning for med- ical imaging. Breakthrough studies in chest radiograph diagnosis, skin and mammographic lesion classification, and diabetic retinopathy detection have relied on hundreds of thousands of labeled training examples [1–4].
    [Show full text]
  • Visual Learning with Weak Supervision
    Visual Learning with Weak Supervision Safa Cicek Committee: Prof. Stefano Soatto, Chair Prof. Lieven Vandenberghe Prof. Paulo Tabuada Prof. Guy Van den Broeck PHD Defense January 2021 UCLA ELECTRICAL AND COMPUTER ENGINEERING Table of Contents Introduction SaaS: Speed as a Supervisor for Semi-supervised Learning [1] Input and Weight Space Smoothing for Semi-supervised Learning [2] Unsupervised Domain Adaptation via Regularized Conditional Alignment [3] Disentangled Image Generation for Unsupervised Domain Adaptation [4] Spatial Class Distribution Shift in Unsupervised Domain Adaptation [5] Learning Topology from Synthetic Data for Unsupervised Depth Completion [6] Targeted Adversarial Perturbations for Monocular Depth Prediction [7] Concluding Remarks [1] Cicek, Safa, Alhussein Fawzi, and Stefano Soatto. Saas: Speed as a supervisor for semi-supervised learning. Proceedings of the European Conference on Computer Vision (ECCV). 2018. [2] Cicek, Safa, and Stefano Soatto. Input and Weight Space Smoothing for Semi-supervised Learning. Proceedings of the IEEE International Conference on Computer Vision (ICCV) Workshops. 2019. [3] Cicek, Safa, and Stefano Soatto. Unsupervised domain adaptation via regularized conditional alignment. Proceedings of the IEEE International Conference on Computer Vision (ICCV). 2019. [4] Cicek, Safa, Zhaowen Wang, Hailin Jin, Stefano Soatto, Generative Feature Disentangling for Unsupervised Domain Adaptation. Proceedings of the European Conference on Computer Vision (ECCV) Workshops. (2020). [5] Cicek, Safa, Ning Xu, Zhaowen Wang, Hailin Jin, Stefano Soatto, Spatial Class Distribution Shift in Unsupervised Domain Adaptation. Asian Conference on Computer Vision (ACCV). 2020. [6] Wong Alex, Safa Cicek, Stefano Soatto, Learning Topology from Synthetic Data for Unsupervised Depth Completion, IEEE Robotics and Automation Letters (RAL). 2021. [7] Wong Alex, Safa Cicek, Stefano Soatto, Targeted Adversarial Perturbations for Monocular Depth Prediction.
    [Show full text]
  • Natural Language Processing and Weak Supervision
    Natural language processing and weak supervision L´eonBottou COS 424 { 4/27/2010 Introduction Natural language processing \from scratch" { Natural language processing systems are heavily engineered. { How much engineering can we avoid by using more data ? { Work by Ronan Collobert, Jason Weston, and the NEC team. Summary { Natural language processing { Embeddings and models { Lots of unlabeled data { Task dependent hacks 2/43 COS 424 { 4/27/2010 I. Natural language processing The Goal We want to have a conversation with our computer . still a long way before HAL 9000 . Convert a piece of English into a computer-friendly data structure How to measure if the computer \understands" something? 4/43 COS 424 { 4/27/2010 Natural Language Processing Tasks Intermediate steps to reach the goal? Part-Of-Speech Tagging (POS): syntactic roles (noun, adverb...) Chunking (CHUNK): syntactic constituents (noun phrase, verb phrase...) Name Entity Recognition (NER): person/company/location... Semantic Role Labeling (SRL): semantic role [John]ARG0 [ate]REL [the apple]ARG1 [in the garden]ARGM LOC − 5/43 COS 424 { 4/27/2010 NLP Benchmarks Datasets: ? POS, CHUNK, SRL: WSJ( up to 1M labeled words) ? NER: Reuters( 200K labeled≈ words) ≈ System Accuracy System F1 Shen, 2007 97.33% Shen, 2005 95.23% Toutanova, 2003 97.24% Sha, 2003 94.29% Gimenez, 2004 97.16% Kudoh, 2001 93.91% (a) POS: As in (Toutanova, 2003) (b) CHUNK: CoNLL 2000 System F1 System F1 Ando, 2005 89.31% Koomen, 2005 77.92% Florian, 2003 88.76% Pradhan, 2005 77.30% Kudoh, 2001 88.31% Haghighi, 2005
    [Show full text]
  • Towards a Perceptually Relevant Control Space
    Music sound synthesis using machine learning: Towards a perceptually relevant control space Fanny ROCHE PhD Defense 29 September 2020 Supervisor: Laurent GIRIN Co-supervisors: Thomas HUEBER Co-supervisors: Maëva GARNIER Co-supervisors: Samuel LIMIER Fanny ROCHE - PhD Defense - 29 September 2020 1 / 48 Context and Objectives Context and Objectives Fanny ROCHE - PhD Defense - 29 September 2020 2 / 48 Context and Objectives Context and Objectives Fanny ROCHE - PhD Defense - 29 September 2020 2 / 48 Context and Objectives Context and Objectives Fanny ROCHE - PhD Defense - 29 September 2020 2 / 48 Processed recordings ! e.g. sampling, wavetable, granular, ... + computationally efficient − very memory consuming Spectral modeling ! e.g. additive, subtractive, ... + close to human sound perception − numerous & very specific Physical modeling ! e.g. solving wave equations, modal synthesis, ... + physically meaningful controls − very specific sounds Context and Objectives Context and Objectives Music sound synthesis Abstract algorithms ! e.g. FM, waveshaping, ... + rich sounds − complex parameters Fanny ROCHE - PhD Defense - 29 September 2020 3 / 48 Spectral modeling ! e.g. additive, subtractive, ... + close to human sound perception − numerous & very specific Physical modeling ! e.g. solving wave equations, modal synthesis, ... + physically meaningful controls − very specific sounds Context and Objectives Context and Objectives Music sound synthesis Abstract algorithms ! e.g. FM, waveshaping, ... + rich sounds − complex parameters Processed recordings ! e.g. sampling, wavetable, granular, ... + computationally efficient − very memory consuming Fanny ROCHE - PhD Defense - 29 September 2020 3 / 48 Physical modeling ! e.g. solving wave equations, modal synthesis, ... + physically meaningful controls − very specific sounds Context and Objectives Context and Objectives Music sound synthesis Abstract algorithms ! e.g. FM, waveshaping, ... + rich sounds − complex parameters Processed recordings ! e.g.
    [Show full text]
  • Self-Supervised Collaborative Learning for Medical Image Synthesis
    The Thirty-Fourth AAAI Conference on Artificial Intelligence (AAAI-20) Auto-GAN: Self-Supervised Collaborative Learning for Medical Image Synthesis Bing Cao,1 Han Zhang,2∗ Nannan Wang,1∗ Xinbo Gao,1 Dinggang Shen2∗ 1State Key Laboratory of Integrated Services Networks, Xidian University, Xi’an, China 2Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, USA Abstract tors desire as many different modalities as possible to make a joint and more accurate decision, certain limitations (e.g., In various clinical scenarios, medical image is crucial in dis- restricted medical conditions, inadequate scanning time, and ease diagnosis and treatment. Different modalities of medical images provide complementary information and jointly helps cost/spending/resource control) could result in sacrificing doctors to make accurate clinical decision. However, due to some imaging modalities to have the most important one clinical and practical restrictions, certain imaging modalities done. For examples, low-resolution (e.g., thick-sliced) imag- may be unavailable nor complete. To impute missing data ing modalities are always combined with another single with adequate clinical accuracy, here we propose a frame- high-resolution imaging modality to save scanning time and work called self-supervised collaborative learning to syn- the low-resolution images can be misleading due to its par- thesize missing modality for medical images. The proposed tial coverage. CT is notorious for poor soft-tissue contrast method comprehensively utilize all available information cor- and radiation exposure, but can better quantify tissue’s phys- related to the target modality from multi-source-modality im- ical density. MRI is non-invasive and safer with better con- ages to generate any missing modality in a single model.
    [Show full text]
  • Weakly Supervised Text Classification with Naturally Annotated Resource
    Automated Knowledge Base Construction (2021) Conference paper NatCat: Weakly Supervised Text Classification with Naturally Annotated Resources Zewei Chu [email protected] University of Chicago, Chicago, IL 60637, USA Karl Stratos [email protected] Rutgers University, Piscataway, NJ 08854, USA Kevin Gimpel [email protected] Toyota Technological Institute at Chicago, Chicago, IL 60637, USA Abstract We describe NatCat, a large-scale resource for text classification constructed from three data sources: Wikipedia, Stack Exchange, and Reddit. NatCat consists of document- category pairs derived from manual curation that occurs naturally within online communities. To demonstrate its usefulness, we build general purpose text classifiers by training on NatCat and evaluate them on a suite of 11 text classification tasks (CatEval), reporting large improvements compared to prior work. We benchmark different modeling choices and resource combinations and show how tasks benefit from particular NatCat data sources.1 1. Introduction Websites with community contributed content contain ample knowledge for natural lan- guage processing. In this paper, we seek to improve text classification by leveraging this knowledge in the form of texts paired with natural category annotations. In particular, we present NatCat, a large-scale resource constructed automatically from three data sources: Wikipedia, Stack Exchange,2 and Reddit.3 With rich knowledge about text and categories, we show that NatCat is a useful resource for text classification. To demonstrate the usefulness of NatCat, we use it to train models that compute a score for any document-category pair. As a result, we get weakly supervised (i.e., \dataless"; Chang et al., 2008) text classifiers that can be used off-the-shelf to produce interpretable and relevant topics for any document.
    [Show full text]
  • Weakly Supervised Short Text Categorization Using World Knowledge
    Weakly Supervised Short Text Categorization Using World Knowledge Rima T¨urker1;2[0000−0001−8332−7241], Lei Zhang1[0000−0001−8184−9353], Mehwish Alam1;2[0000−0002−7867−6612], and Harald Sack1;2[0000−0001−7069−9804] 1 FIZ Karlsruhe { Leibniz Institute for Information Infrastructure, Germany 2 Karlsruhe Institute of Technology, Institute AIFB, Germany [email protected] [email protected] Abstract. Short text categorization is an important task in many NLP applications, such as sentiment analysis, news feed categorization, etc. Due to the sparsity and shortness of the text, many traditional classifi- cation models perform poorly if they are directly applied to short text. Moreover, supervised approaches require large amounts of manually la- beled data, which is a costly, labor intensive, and time-consuming task. This paper proposes a weakly supervised short text categorization ap- proach, which does not require any manually labeled data. The proposed model consists of two main modules: (1) a data labeling module, which leverages an external Knowledge Base (KB) to compute probabilistic labels for a given unlabeled training data set, and (2) a classification model based on a Wide & Deep learning approach. The effectiveness of the proposed method is validated via evaluation on multiple datasets. The experimental results show that the proposed approach outperforms unsupervised state-of-the-art classification approaches and achieves com- parable performance to supervised approaches. Keywords: Short Text Categorization · Weakly Supervised Short Text Categorization · Wide & Deep Model. 1 Introduction Due to rapid growth of the Web content, short text data such as search snippets, news feeds, short messages, etc.
    [Show full text]
  • Strength from Weakness: Fast Learning Using Weak Supervision
    Strength from Weakness: Fast Learning Using Weak Supervision Joshua Robinson 1 Stefanie Jegelka 1 Suvrit Sra 1 Abstract plex model on a related, large data task, and to then uses We study generalization properties of weakly su- the learned features for finetuning on the small-data target pervised learning, that is, learning where only a task (Girshick et al., 2014; Donahue et al., 2014; Zeiler & few “strong” labels (the actual target for predic- Fergus, 2014; Sun et al., 2017). Numerous approaches to tion) are present but many more “weak” labels are weakly supervised learning have succeeded in a variety of available. In particular, we show that pretraining tasks; beyond computer vision (Oquab et al., 2015; Durand using weak labels and finetuning using strong can et al., 2017; Carreira & Zisserman, 2017; Fries et al., 2019). accelerate the learning rate for the strong task to Examples include clinical text classification (Wang et al., 2019), sentiment analysis (Medlock & Briscoe, 2007), so- the fast rate of (1/n), where n is the number of strongly labeledO data points. This acceleration cial media content tagging (Mahajan et al., 2018) and many can happen even if, by itself, the strongly labeled others. Weak supervision is also closely related to unsu- pervised learning methods such as complementary and con- data admits only the slower (1/√n) rate. The acceleration depends continuouslyO on the number trastive learning (Xu et al., 2019; Chen & Batmanghelich, of weak labels available, and on the relation be- 2019; Arora et al., 2019), and particularly to self-supervised tween the two tasks. Our theoretical results are learning (Doersch et al., 2015), where feature maps learned reflected empirically across a range of tasks and via supervised training on artificially constructed tasks have illustrate how weak labels speed up learning on been found to even outperform ImageNet learned features on the strong task.
    [Show full text]
  • Snuba: Automating Weak Supervision to Label Training Data
    Snuba: Automating Weak Supervision to Label Training Data Paroma Varma Christopher Re´ Stanford University Stanford University [email protected] [email protected] ABSTRACT benign, malignant if area > 210.8: As deep learning models are applied to increasingly diverse return False Label if area < 150: Aggregator problems, a key bottleneck is gathering enough high-quality return Abstain training labels tailored to each task. Users therefore turn Labeled Data 25% benign, 75%benign, ... ?, ?, ?, ? if perim > 120: to weak supervision, relying on imperfect sources of labels return True like pattern matching and user-defined heuristics. Unfor- if perim > 80 Terminate? return Abstain tunately, users have to design these sources for each task. Heuristic Generation Training Labels This process can be time consuming and expensive: domain Unlabeled Data Snuba experts often perform repetitive steps like guessing optimal numerical thresholds and developing informative text pat- Figure 1: Snuba uses a small labeled and a large unlabeled terns. To address these challenges, we present Snuba,a dataset to iteratively generate heuristics. It uses existing la- system to automatically generate heuristics using a small bel aggregators to assign training labels to the large dataset. labeled dataset to assign training labels to a large, unla- beled dataset in the weak supervision setting. Snuba gen- pervision, or methods that can assign noisy training labels erates heuristics that each labels the subset of the data it to unlabeled data, like crowdsourcing [9, 22, 60], distant su- is accurate for, and iteratively repeats this process until the pervision [8,32], and user-defined heuristics [38,39,50]. Over heuristics together label a large portion of the unlabeled the past few years, we have been part of the broader effort data.
    [Show full text]
  • Multi-Modal Weakly Supervised Dense Event Captioning
    Watch, Listen and Tell: Multi-modal Weakly Supervised Dense Event Captioning Tanzila Rahman1,2 Bicheng Xu1,2 Leonid Sigal1,2,3 1University of British Columbia 2Vector Institute for AI 3Canada CIFAR AI Chair {trahman8, bichengx, lsigal}@cs.ubc.ca Abstract Input Video with Audio Track Dense Event Captions A group of friends sit Multi-modal learning, particularly among imaging and together in a living room linguistic modalities, has made amazing strides in many talking. high-level fundamental visual understanding problems, .. ranging from language grounding to dense event caption- One of the friends plays . a song on his guitar and ing. However, much of the research has been limited to the others listen. approaches that either do not take audio corresponding to .. Time video into account at all, or those that model the audio- . One friend falls back visual correlations in service of sound or sound source lo- into the chair in . amazement after the calization. In this paper, we present the evidence, that audio song . signals can carry surprising amount of information when .. it comes to high-level visual-lingual tasks. Specifically, The two friends congratulate the we focus on the problem of weakly-supervised dense event man for his guitar performance and captioning in videos and show that audio on its own can shake hands. nearly rival performance of a state-of-the-art visual model and, combined with video, can improve on the state-of-the- Figure 1: Multi-modal Dense Event Captioning. Illus- art performance. Extensive experiments on the ActivityNet tration of our problem definition, where we use both audio Captions dataset show that our proposed multi-modal ap- features and visual information to generate the dense cap- proach outperforms state-of-the-art unimodal methods, as tions for a video in a weakly supervised manner.
    [Show full text]