No abstract

We investigate the recently developed Bidirectional Encoder Representations from Transformers (BERT) model (Devlin et al., 2018) for the hyperpartisan news detection task. Using a subset of hand-labeled articles from Se-mEval as a validation set, we test the performance of different parameters for BERT models. We find that accuracy from two different BERT models using different proportions of the articles is consistently high, with our bestperforming model on the validation set achieving 85% accuracy and the best-performing model on the test set achieving 77%. We further determined that our model exhibits strong consistency, labeling independent slices of the same article identically. Finally, we find that randomizing the order of word pieces dramatically reduces validation accuracy (to approximately 60%), but that shuffling groups of four or more word pieces maintains an accuracy of about 80%, indicating the model mainly gains value from local context.

Algorithm performance in supervised learning is a combination of memorization, generalization, and luck. By estimating how much information an algorithm can memorize from a dataset, we can set a lower bound on the amount of performance due to other factors such as generalization and luck. With this goal in mind, we introduce the Labeling Distribution Matrix (LDM) as a tool for estimating the capacity of learning algorithms. The method attempts to characterize the diversity of possible outputs by an algorithm for different training datasets, using this to measure algorithm flexibility and responsiveness to data. We test the method on several supervised learning algorithms, and find that while the results are not conclusive, the LDM does allow us to gain potentially valuable insight into the prediction behavior of algorithms. We also introduce the Label Recorder as an additional tool for estimating algorithm capacity, with more promising initial results.

We present an information-theoretic framework for understanding overfitting and underfitting in machine learning and prove the formal undecidability of determining whether an arbitrary classification algorithm will overfit a dataset. Measuring algorithm capacity via the information transferred from datasets to models, we consider mismatches between algorithm capacities and datasets to provide a signature for when a model can overfit or underfit a dataset. We present results upperbounding algorithm capacity, establish its relationship to quantities in the algorithmic search framework for machine learning, and relate our work to recent information-theoretic approaches to generalization.