Comparing Test Sets with Item Response Theory

Recent years have seen numerous NLP datasets introduced to evaluate the performance of fine-tuned models on natural language understanding tasks. Recent results from large pretrained models, though, show that many of these datasets are largely saturated and unlikely to be able to detect further progress. What kind of datasets are still effective at discriminating among strong models, and what kind of datasets should we expect to be able to detect future improvements? To measure this uniformly across datasets, we draw on Item Response Theory and evaluate 29 datasets using predictions from 18 pretrained Transformer models on individual test examples. We find that Quoref, HellaSwag, and MC-TACO are best suited for distinguishing among state-of-the-art models, while SNLI, MNLI, and CommitmentBank seem to be saturated for current strong models. We also observe span selection task format, which is used for QA datasets like QAMR or SQuAD2.0, is effective in differentiating between strong and weak models.

True Few-Shot Learning with Language Models

Pretrained language models (LMs) perform well on many tasks even when learning from a few examples, but prior work uses many held-out examples to tune various aspects of learning, such as hyperparameters, training objectives, and natural language templates ("prompts"). Here, we evaluate the few-shot ability of LMs when such held-out examples are unavailable, a setting we call true few-shot learning. We test two model selection criteria, cross-validation and minimum description length, for choosing LM prompts and hyperparameters in the true few-shot setting. On average, both marginally outperform random selection and greatly underperform selection based on held-out examples. Moreover, selection criteria often prefer models that perform significantly worse than randomly-selected ones. We find similar results even when taking into account our uncertainty in a model's true performance during selection, as well as when varying the amount of computation and number of examples used for selection. Overall, our findings suggest that prior work significantly overestimated the true few-shot ability of LMs given the difficulty of few-shot model selection.

KLUE: Korean Language Understanding Evaluation

We introduce Korean Language Understanding Evaluation (KLUE) benchmark. KLUE is a collection of 8 Korean natural language understanding (NLU) tasks, including Topic Classification, SemanticTextual Similarity, Natural Language Inference, Named Entity Recognition, Relation Extraction, Dependency Parsing, Machine Reading Comprehension, and Dialogue State Tracking. We build all of the tasks from scratch from diverse source corpora while respecting copyrights, to ensure accessibility for anyone without any restrictions. With ethical considerations in mind, we carefully design annotation protocols. Along with the benchmark tasks and data, we provide suitable evaluation metrics and fine-tuning recipes for pretrained language models for each task. We furthermore release the pretrained language models (PLM), KLUE-BERT and KLUE-RoBERTa, to help reproducing baseline models on KLUE and thereby facilitate future research. We make a few interesting observations from the preliminary experiments using the proposed KLUE benchmark suite, already demonstrating the usefulness of this new benchmark suite. First, we find KLUE-RoBERTa-large outperforms other baselines, including multilingual PLMs and existing open-source Korean PLMs. Second, we see minimal degradation in performance even when we replace personally identifiable information from the pretraining corpus, suggesting that privacy and NLU capability are not at odds with each other. Lastly, we find that using BPE tokenization in combination with morpheme-level pre-tokenization is effective in tasks involving morpheme-level tagging, detection and generation. In addition to accelerating Korean NLP research, our comprehensive documentation on creating KLUE will facilitate creating similar resources for other languages in the future. KLUE is available at <a class="link-external link-https" href="https://klue-benchmark.com/">this URL</a>.

Future is not One-dimensional: Graph Modeling based Complex Event Schema Induction for Event Prediction

Event schemas encode knowledge of stereotypical structures of events and their connections. As events unfold, schemas are crucial to act as a scaffolding. Previous work on event schema induction either focuses on atomic events or linear temporal event sequences, ignoring the interplay between events via arguments and argument relations. We introduce the concept of Temporal Complex Event Schema: a graph-based schema representation that encompasses events, arguments, temporal connections and argument relations. Additionally, we propose a Temporal Event Graph Model that models the emergence of event instances following the temporal complex event schema. To build and evaluate such schemas, we release a new schema learning corpus containing 6,399 documents accompanied with event graphs, and manually constructed gold schemas. Intrinsic evaluation by schema matching and instance graph perplexity, prove the superior quality of our probabilistic graph schema library compared to linear representations. Extrinsic evaluation on schema-guided event prediction further demonstrates the predictive power of our event graph model, significantly surpassing human schemas and baselines by more than 17.8% on HITS@1.

NaturalProofs: Mathematical Theorem Proving in Natural Language

Understanding and creating mathematics using natural mathematical language - the mixture of symbolic and natural language used by humans - is a challenging and important problem for driving progress in machine learning. As a step in this direction, we develop NaturalProofs, a large-scale dataset of mathematical statements and their proofs, written in natural mathematical language. Using NaturalProofs, we propose a mathematical reference retrieval task that tests a system's ability to determine the key results that appear in a proof. Large-scale sequence models excel at this task compared to classical information retrieval techniques, and benefit from language pretraining, yet their performance leaves substantial room for improvement. NaturalProofs opens many possibilities for future research on challenging mathematical tasks.

Rissanen Data Analysis: Examining Dataset Characteristics via Description Length

We introduce a method to determine if a certain capability helps to achieve an accurate model of given data. We view labels as being generated from the inputs by a program composed of subroutines with different capabilities, and we posit that a subroutine is useful if and only if the minimal program that invokes it is shorter than the one that does not. Since minimum program length is uncomputable, we instead estimate the labels' minimum description length (MDL) as a proxy, giving us a theoretically-grounded method for analyzing dataset characteristics. We call the method Rissanen Data Analysis (RDA) after the father of MDL, and we showcase its applicability on a wide variety of settings in NLP, ranging from evaluating the utility of generating subquestions before answering a question, to analyzing the value of rationales and explanations, to investigating the importance of different parts of speech, and uncovering dataset gender bias.

Online hyperparameter optimization by real-time recurrent learning

Conventional hyperparameter optimization methods are computationally intensive and hard to generalize to scenarios that require dynamically adapting hyperparameters, such as life-long learning. Here, we propose an online hyperparameter optimization algorithm that is asymptotically exact and computationally tractable, both theoretically and practically. Our framework takes advantage of the analogy between hyperparameter optimization and parameter learning in recurrent neural networks (RNNs). It adapts a well-studied family of online learning algorithms for RNNs to tune hyperparameters and network parameters simultaneously, without repeatedly rolling out iterative optimization. This procedure yields systematically better generalization performance compared to standard methods, at a fraction of wallclock time.

Self-Supervised Equivariant Scene Synthesis from Video

We propose a self-supervised framework to learn scene representations from video that are automatically delineated into background, characters, and their animations. Our method capitalizes on moving characters being equivariant with respect to their transformation across frames and the background being constant with respect to that same transformation. After training, we can manipulate image encodings in real time to create unseen combinations of the delineated components. As far as we know, we are the first method to perform unsupervised extraction and synthesis of interpretable background, character, and animation. We demonstrate results on three datasets: Moving MNIST with backgrounds, 2D video game sprites, and Fashion Modeling.

Catastrophic Fisher Explosion: Early Phase Fisher Matrix Impacts Generalization

The early phase of training has been shown to be important in two ways for deep neural networks. First, the degree of regularization in this phase significantly impacts the final generalization. Second, it is accompanied by a rapid change in the local loss curvature influenced by regularization choices. Connecting these two findings, we show that stochastic gradient descent (SGD) implicitly penalizes the trace of the Fisher Information Matrix (FIM) from the beginning of training. We argue it is an implicit regularizer in SGD by showing that explicitly penalizing the trace of the FIM can significantly improve generalization. We further show that the early value of the trace of the FIM correlates strongly with the final generalization. We highlight that in the absence of implicit or explicit regularization, the trace of the FIM can increase to a large value early in training, to which we refer as catastrophic Fisher explosion. Finally, to gain insight into the regularization effect of penalizing the trace of the FIM, we show that 1) it limits memorization by reducing the learning speed of examples with noisy labels more than that of the clean examples, and 2) trajectories with a low initial trace of the FIM end in flat minima, which are commonly associated with good generalization.