Tracking progress in machine learning has become increasingly difficult with the recent explosion in the number of papers. In this paper, we present AxCell, an automatic machine learning pipeline for extracting results from papers. AxCell uses several novel components, including a table segmentation subtask, to learn relevant structural knowledge that aids extraction. When compared with existing methods, our approach significantly improves the state of the art for results extraction. We also release a structured, annotated dataset for training models for results extraction, and a dataset for evaluating the performance of models on this task. Lastly, we show the viability of our approach enables it to be used for semi-automated results extraction in production, suggesting our improvements make this task practically viable for the first time. Code is available on GitHub.
While recent efforts have shown that neural text processing models are vulnerable to adversarial examples, comparatively little attention has been paid to explicitly characterize their effectiveness. To overcome this, we present analytical insights into the word frequency characteristics of word-level adversarial examples for neural text classification models. We show that adversarial attacks against CNN-, LSTM- and Transformer-based classification models perform token substitutions that are identifiable through word frequency differences between replaced words and their substitutions. Based on these findings, we propose frequency-guided word substitutions (FGWS) as a simple algorithm for the automatic detection of adversarially perturbed textual sequences. FGWS exploits the word frequency properties of adversarial word substitutions, and we assess its suitability for the automatic detection of adversarial examples generated from the SST-2 and IMDb sentiment datasets. Our method provides promising results by accurately detecting adversarial examples, with $F_1$ detection scores of up to 93.7% on adversarial examples against BERT-based classification models. We compare our approach against baseline detection approaches as well as a recently proposed perturbation discrimination framework, and show that we outperform existing approaches by up to 15.1% $F_1$ in our experiments.
Machine reading comprehension (MRC) has received considerable attention in natural language processing over the past few years. However, the conventional task design of MRC lacks the explainability beyond the model interpretation, i.e., the internal mechanics of the model cannot be explained in human terms. To this end, this position paper provides a theoretical basis for the design of MRC based on psychology and psychometrics and summarizes it in terms of the requirements for explainable MRC. We conclude that future datasets should (i) evaluate the capability of the model for constructing a coherent and grounded representation to understand context-dependent situations and (ii) ensure substantive validity by improving the question quality and by formulating a white-box task.
Current reading comprehension models generalise well to in-distribution test sets, yet perform poorly on adversarially selected inputs. Most prior work on adversarial inputs studies oversensitivity: semantically invariant text perturbations that cause a model's prediction to change when it should not. In this work we focus on the complementary problem: excessive prediction undersensitivity, where input text is meaningfully changed but the model's prediction does not, even though it should. We formulate a noisy adversarial attack which searches among semantic variations of the question for which a model erroneously predicts the same answer, and with even higher probability. Despite comprising unanswerable questions, both SQuAD2.0 and NewsQA models are vulnerable to this attack. This indicates that although accurate, models tend to rely on spurious patterns and do not fully consider the information specified in a question. We experiment with data augmentation and adversarial training as defences, and find that both substantially decrease vulnerability to attacks on held out data, as well as held out attack spaces. Addressing undersensitivity also improves results on AddSent and AddOneSent, and models furthermore generalise better when facing train/evaluation distribution mismatch: they are less prone to overly rely on predictive cues present only in the training set, and outperform a conventional model by as much as 10.9% F1.
Innovations in annotation methodology have been a propellant for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation approach and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalisation to data collected without a model. We find that training on adversarially collected samples leads to strong generalisation to non-adversarially collected datasets, yet with progressive deterioration as the model-in-the-loop strength increases. Furthermore we find that stronger models can still learn from datasets collected with substantially weaker models in the loop: When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 36.0F1 on questions that it cannot answer when trained on SQuAD - only marginally lower than when trained on data collected using RoBERTa itself.
Existing analysis work in machine reading comprehension (MRC) is largely concerned with evaluating the capabilities of systems. However, the capabilities of datasets are not assessed for benchmarking language understanding precisely. We propose a semi-automated, ablation-based methodology for this challenge; By checking whether questions can be solved even after removing features associated with a skill requisite for language understanding, we evaluate to what degree the questions do not require the skill. Experiments on 10 datasets (e.g., CoQA, SQuAD v2.0, and RACE) with a strong baseline model show that, for example, the relative scores of a baseline model provided with content words only and with shuffled sentence words in the context are on average 89.2% and 78.5% of the original score, respectively. These results suggest that most of the questions already answered correctly by the model do not necessarily require grammatical and complex reasoning. For precise benchmarking, MRC datasets will need to take extra care in their design to ensure that questions can correctly evaluate the intended skills.
Recent studies revealed that reading comprehension (RC) systems learn to exploit annotation artifacts and other biases in current datasets. This allows systems to "cheat" by employing simple heuristics to answer questions, e.g. by relying on semantic type consistency. This means that current datasets are not well-suited to evaluate RC systems. To address this issue, we introduce RC-QED, a new RC task that requires giving not only the correct answer to a question, but also the reasoning employed for arriving at this answer. For this, we release a large benchmark dataset consisting of 12,000 answers and corresponding reasoning in form of natural language derivations. Experiments show that our benchmark is robust to simple heuristics and challenging for state-of-the-art neural path ranking approaches.
We present an approach for automatic extraction of measured values from the astrophysical literature, using the Hubble constant for our pilot study. Our rules-based model -- a classical technique in natural language processing -- has successfully extracted 298 measurements of the Hubble constant, with uncertainties, from the 208,541 available arXiv astrophysics papers. We have also created an artificial neural network classifier to identify papers which report novel measurements. This classifier is applied to the available arXiv data, and is demonstrated to work well in identifying papers which are reporting new measurements. From the analysis of our results we find that reporting measurements with uncertainties and the correct units is critical information to identify novel measurements in free text. Our results correctly highlight the current tension for measurements of the Hubble constant and recover the $3.5\sigma$ discrepancy -- demonstrating that the tool presented in this paper is useful for meta-studies of astrophysical measurements from a large number of publications, and showing the potential to generalise this technique to other areas.
Like all sub-fields of machine learning Bayesian Deep Learning is driven by empirical validation of its theoretical proposals. Given the many aspects of an experiment it is always possible that minor or even major experimental flaws can slip by both authors and reviewers. One of the most popular experiments used to evaluate approximate inference techniques is the regression experiment on UCI datasets. However, in this experiment, models which have been trained to convergence have often been compared with baselines trained only for a fixed number of iterations. We find that a well-established baseline, Monte Carlo dropout, when evaluated under the same experimental settings shows significant improvements. In fact, the baseline outperforms or performs competitively with methods that claimed to be superior to the very same baseline method when they were introduced. Hence, by exposing this flaw in experimental procedure, we highlight the importance of using identical experimental setups to evaluate, compare, and benchmark methods in Bayesian Deep Learning.
Electronic health records (EHR) are increasingly being used for constructing disease risk prediction models. Feature engineering in EHR data however is challenging due to their highly dimensional and heterogeneous nature. Low-dimensional representations of EHR data can potentially mitigate these challenges. In this paper, we use global vectors (GloVe) to learn word embeddings for diagnoses and procedures recorded using 13 million ontology terms across 2.7 million hospitalisations in national UK EHR. We demonstrate the utility of these embeddings by evaluating their performance in identifying patients which are at higher risk of being hospitalised for congestive heart failure. Our findings indicate that embeddings can enable the creation of robust EHR-derived disease risk prediction models and address some the limitations associated with manual clinical feature engineering.