We present MLRG Deep Curvature suite, a PyTorch-based, open-source package for analysis and visualisation of neural network curvature and loss landscape. Despite of providing rich information into properties of neural network and useful for a various designed tasks, curvature information is still not made sufficient use for various reasons, and our method aims to bridge this gap. We present a primer, including its main practical desiderata and common misconceptions, of \textit{Lanczos algorithm}, the theoretical backbone of our package, and present a series of examples based on synthetic toy examples and realistic modern neural networks tested on CIFAR datasets, and show the superiority of our package against existing competing approaches for the similar purposes.
Empirical research in Natural Language Processing (NLP) has adopted a narrow set of principles for assessing hypotheses, relying mainly on p-value computation, which suffers from several known issues. While alternative proposals have been well-debated and adopted in other fields, they remain rarely discussed or used within the NLP community. We address this gap by contrasting various hypothesis assessment techniques, especially those not commonly used in the field (such as evaluations based on Bayesian inference). Since these statistical techniques differ in the hypotheses they can support, we argue that practitioners should first decide their target hypothesis before choosing an assessment method. This is crucial because common fallacies, misconceptions, and misinterpretation surrounding hypothesis assessment methods often stem from a discrepancy between what one would like to claim versus what the method used actually assesses. Our survey reveals that these issues are omnipresent in the NLP research community. As a step forward, we provide best practices and guidelines tailored to NLP research, as well as an easy-to-use package called 'HyBayes' for Bayesian assessment of hypotheses, complementing existing tools.
Discounted reinforcement learning is fundamentally incompatible with function approximation for control in continuing tasks. It is not an optimization problem in its usual formulation, so when using function approximation there is no optimal policy. We substantiate these claims, then go on to address some misconceptions about discounting and its connection to the average reward formulation. We encourage researchers to adopt rigorous optimization approaches, such as maximizing average reward, for reinforcement learning in continuing tasks.
Structural roles define sets of structurally similar nodes that are more similar to nodes inside the set than outside, whereas communities define sets of nodes with more connections inside the set than outside. Roles based on structural similarity and communities based on proximity are fundamentally different but important complementary notions. Recently, the notion of structural roles has become increasingly important and has gained a lot of attention due to the proliferation of work on learning representations (node/edge embeddings) from graphs that preserve the notion of roles. Unfortunately, recent work has sometimes confused the notion of structural roles and communities (based on proximity) leading to misleading or incorrect claims about the capabilities of network embedding methods. As such, this paper seeks to clarify the misconceptions and key differences between structural roles and communities, and formalize the general mechanisms (e.g., random walks, feature diffusion) that give rise to community or role-based structural embeddings. We theoretically prove that embedding methods based on these mechanisms result in either community or role-based structural embeddings. These mechanisms are typically easy to identify and can help researchers quickly determine whether a method preserves community or role-based embeddings. Furthermore, they also serve as a basis for developing new and improved methods for community or role-based structural embeddings. Finally, we analyze and discuss applications and data characteristics where community or role-based embeddings are most appropriate.
The misclassification error distance and the adjusted Rand index are two of the most commonly used criteria to evaluate the performance of clustering algorithms. This paper provides an in-depth comparison of the two criteria, aimed to better understand exactly what they measure, their properties and their differences. Starting from their population origins, the investigation includes many data analysis examples and the study of particular cases in great detail. An exhaustive simulation study allows inspecting the criteria distributions and reveals some previous misconceptions.
Cross-lingual word embeddings (CLEs) enable multilingual modeling of meaning and facilitate cross-lingual transfer of NLP models. Despite their ubiquitous usage in downstream tasks, recent increasingly popular projection-based CLE models are almost exclusively evaluated on a single task only: bilingual lexicon induction (BLI). Even BLI evaluations vary greatly, hindering our ability to correctly interpret performance and properties of different CLE models. In this work, we make the first step towards a comprehensive evaluation of cross-lingual word embeddings. We thoroughly evaluate both supervised and unsupervised CLE models on a large number of language pairs in the BLI task and three downstream tasks, providing new insights concerning the ability of cutting-edge CLE models to support cross-lingual NLP. We empirically demonstrate that the performance of CLE models largely depends on the task at hand and that optimizing CLE models for BLI can result in deteriorated downstream performance. We indicate the most robust supervised and unsupervised CLE models and emphasize the need to reassess existing baselines, which still display competitive performance across the board. We hope that our work will catalyze further work on CLE evaluation and model analysis.
Community detection in networks is one of the most popular topics of modern network science. Communities, or clusters, are usually groups of vertices having higher probability of being connected to each other than to members of other groups, though other patterns are possible. Identifying communities is an ill-defined problem. There are no universal protocols on the fundamental ingredients, like the definition of community itself, nor on other crucial issues, like the validation of algorithms and the comparison of their performances. This has generated a number of confusions and misconceptions, which undermine the progress in the field. We offer a guided tour through the main aspects of the problem. We also point out strengths and weaknesses of popular methods, and give directions to their use.
In modern computer science education, massive open online courses (MOOCs) log thousands of hours of data about how students solve coding challenges. Being so rich in data, these platforms have garnered the interest of the machine learning community, with many new algorithms attempting to autonomously provide feedback to help future students learn. But what about those first hundred thousand students? In most educational contexts (i.e. classrooms), assignments do not have enough historical data for supervised learning. In this paper, we introduce a human-in-the-loop "rubric sampling" approach to tackle the "zero shot" feedback challenge. We are able to provide autonomous feedback for the first students working on an introductory programming assignment with accuracy that substantially outperforms data-hungry algorithms and approaches human level fidelity. Rubric sampling requires minimal teacher effort, can associate feedback with specific parts of a student's solution and can articulate a student's misconceptions in the language of the instructor. Deep learning inference enables rubric sampling to further improve as more assignment specific student data is acquired. We demonstrate our results on a novel dataset from Code.org, the world's largest programming education platform.
Question answering is an important task for autonomous agents and virtual assistants alike and was shown to support the disabled in efficiently navigating an overwhelming environment. Many existing methods focus on observation-based questions, ignoring our ability to seamlessly combine observed content with general knowledge. To understand interactions with a knowledge base, a dataset has been introduced recently and keyword matching techniques were shown to yield compelling results despite being vulnerable to misconceptions due to synonyms and homographs. To address this issue, we develop a learning-based approach which goes straight to the facts via a learned embedding space. We demonstrate state-of-the-art results on the challenging recently introduced fact-based visual question answering dataset, outperforming competing methods by more than 5%.
Two different views on machine learning problem: Applied learning (machine learning with business applications) and Agnostic PAC learning are formalized and compared here. I show that, under some conditions, the theory of PAC Learnable provides a way to solve the Applied learning problem. However, the theory requires to have the training sets so large, that it would make the learning practically useless. I suggest shedding some theoretical misconceptions about learning to make the theory more aligned with the needs and experience of practitioners.
We demonstrate that the concerns expressed by Garcia et al. are misplaced, due to (1) a misreading of our findings in [1]; (2) a widespread failure to examine and present words in support of asserted summary quantities based on word usage frequencies; and (3) a range of misconceptions about word usage frequency, word rank, and expert-constructed word lists. In particular, we show that the English component of our study compares well statistically with two related surveys, that no survey design influence is apparent, and that estimates of measurement error do not explain the positivity biases reported in our work and that of others. We further demonstrate that for the frequency dependence of positivity---of which we explored the nuances in great detail in [1]---Garcia et al. did not perform a reanalysis of our data---they instead carried out an analysis of a different, statistically improper data set and introduced a nonlinearity before performing linear regression.
This report seeks to inform policy makers on the nature and the merit of the arguments for and against the concerns associated with a potential technological singularity. Part I describes the lessons learned from our investigation of the subject, separating the argu-ments of merit from the fallacies and misconceptions that confuse the debate and undermine its rational resolution.
An important, yet largely unstudied, problem in student data analysis is to detect misconceptions from students' responses to open-response questions. Misconception detection enables instructors to deliver more targeted feedback on the misconceptions exhibited by many students in their class, thus improving the quality of instruction. In this paper, we propose a new natural language processing-based framework to detect the common misconceptions among students' textual responses to short-answer questions. We propose a probabilistic model for students' textual responses involving misconceptions and experimentally validate it on a real-world student-response dataset. Experimental results show that our proposed framework excels at classifying whether a response exhibits one or more misconceptions. More importantly, it can also automatically detect the common misconceptions exhibited across responses from multiple students to multiple questions; this property is especially important at large scale, since instructors will no longer need to manually specify all possible misconceptions that students might exhibit.
Learning-based pattern classifiers, including deep networks, have shown impressive performance in several application domains, ranging from computer vision to cybersecurity. However, it has also been shown that adversarial input perturbations carefully crafted either at training or at test time can easily subvert their predictions. The vulnerability of machine learning to such wild patterns (also referred to as adversarial examples), along with the design of suitable countermeasures, have been investigated in the research field of adversarial machine learning. In this work, we provide a thorough overview of the evolution of this research area over the last ten years and beyond, starting from pioneering, earlier work on the security of non-deep learning algorithms up to more recent work aimed to understand the security properties of deep learning algorithms, in the context of computer vision and cybersecurity tasks. We report interesting connections between these apparently-different lines of work, highlighting common misconceptions related to the security evaluation of machine-learning algorithms. We review the main threat models and attacks defined to this end, and discuss the main limitations of current work, along with the corresponding future challenges towards the design of more secure learning algorithms.
Generative adversarial networks (GANs) form a generative modeling approach known for producing appealing samples, but they are notably difficult to train. One common way to tackle this issue has been to propose new formulations of the GAN objective. Yet, surprisingly few studies have looked at optimization methods designed for this adversarial training. In this work, we cast GAN optimization problems in the general variational inequality framework. Tapping into the mathematical programming literature, we counter some common misconceptions about the difficulties of saddle point optimization and propose to extend techniques designed for variational inequalities to the training of GANs. We apply averaging, extrapolation and a computationally cheaper variant that we call extrapolation from the past to the stochastic gradient method (SGD) and Adam.
We investigate the design challenges of constructing effective and efficient neural sequence labeling systems, by reproducing twelve neural sequence labeling models, which include most of the state-of-the-art structures, and conduct a systematic model comparison on three benchmarks (i.e. NER, Chunking, and POS tagging). Misconceptions and inconsistent conclusions in existing literature are examined and clarified under statistical experiments. In the comparison and analysis process, we reach several practical conclusions which can be useful to practitioners.
The lack of comprehensive, high-quality health data in developing nations creates a roadblock for combating the impacts of disease. One key challenge is understanding the health information needs of people in these nations. Without understanding people's everyday needs, concerns, and misconceptions, health organizations and policymakers lack the ability to effectively target education and programming efforts. In this paper, we propose a bottom-up approach that uses search data from individuals to uncover and gain insight into health information needs in Africa. We analyze Bing searches related to HIV/AIDS, malaria, and tuberculosis from all 54 African nations. For each disease, we automatically derive a set of common search themes or topics, revealing a wide-spread interest in various types of information, including disease symptoms, drugs, concerns about breastfeeding, as well as stigma, beliefs in natural cures, and other topics that may be hard to uncover through traditional surveys. We expose the different patterns that emerge in health information needs by demographic groups (age and sex) and country. We also uncover discrepancies in the quality of content returned by search engines to users by topic. Combined, our results suggest that search data can help illuminate health information needs in Africa and inform discussions on health policy and targeted education efforts both on- and offline.