What do Deep Networks like to Read?

Our paper What do Deep Networks like to Read? is available on arXiv!

Here, we uncover artifacts that are encoded in our models, by finetuning a pretrained autoencoder using gradients from a classifier with frozen weights.

(1) We pre-train a Seq2Seq autoencoder to initially adequately reproduce the input text. (2) We pre-train different classification models for diverse tasks. (3) We fine-tune the weights of the autoencoder with the gradients of the classifier with fixed weights.

This rephrases the original sentences and gives us a view into what deep networks like to read. We find differences in how CNNs, LSTMs and Deep Averaging Networks restructure the sentences.

By flipping the labels from the gold class, we force the autoencoder to rephrase the sentence in away that the classifier predicts the „new“ label. This helps us identify if the model has possibly encoded artifacts of the data set.

Abstract: Recent research towards understanding neural networks probes models in a top-down manner, but is only able to identify model tendencies that are known a priori. We propose Susceptibility Identification through Fine-Tuning (SIFT), a novel abstractive method that uncovers a model’s preferences without imposing any prior. By fine-tuning an autoencoder with the gradients from a fixed classifier, we are able to extract propensities that characterize different kinds of classifiers in a bottom-up manner. We further leverage the SIFT architecture to rephrase sentences in order to predict the opposing class of the ground truth label, uncovering potential artifacts encoded in the fixed classification model. We evaluate our method on three diverse tasks with four different models. We contrast the propensities of the models as well as reproduce artifacts reported in the literature.

Demo Paper accepted at EMNLP 2019

I am happy to annouce that our Demo Paper FAMULUS: Interactive Annotation and Feedback Generation for Teaching Diagnostic Reasoning was accepted at EMNLP 2019. We will be presenting our work during the demo session in Hong Kong in November 2019.

Abstract: Our proposed system FAMULUS helps students learn to diagnose based on automatic feedback in virtual patient simulations, and it supports instructors in labeling training data. Diagnosing is an exceptionally difficult skill to obtain but vital for many different professions (e.g., medical doctors, teachers). Previous case simulation systems are limited to multiple-choice questions and thus cannot give constructive individualized feedback on a student’s diagnostic reasoning process. Given initially only limited data, we leverage a (replaceable) NLP model to both support experts in their further data annotation with automatic suggestions, and we provide automatic feedback for students. We argue that because the central model consistently improves, our interactive approach encourages both students and instructors to recurrently use the tool, and thus accelerate the speed of data creation and annotation. We show results from two user studies on diagnostic reasoning in medicine and teacher education and outline how our system can be extended to further use cases.

Update - Best Paper Award at Repl4NLP

Our paper Specializing Distributional Vectors of All Words for Lexical Entailment won the best paper award!

Abstract: Semantic specialization methods fine-tune distributional word vectors using lexical knowledge from external resources (e.g., WordNet) to accentuate a particular relation between words. However, such post-processing methods suffer from limited coverage as they affect only vectors of words seen in the external resources. We present the first postprocessing method that specializes vectors of all vocabulary words – including those unseen in the resources – for the asymmetric relation of lexical entailment (LE) (i.e., hyponymyhypernymy relation). Leveraging a partially LE-specialized distributional space, our POSTLE (i.e., post-specialization for LE) model learns an explicit global specialization function, allowing for specialization of vectors of unseen words, as well as word vectors from other languages via cross-lingual transfer. We capture the function as a deep feedforward neural network: its objective re-scales vector norms to reflect the concept hierarchy while simultaneously attracting hyponymyhypernymy pairs to better reflect semantic similarity. An extended model variant augments the basic architecture with an adversarial discriminator. We demonstrate the usefulness and versatility of POSTLE models with different input distributional spaces in different scenarios (monolingual LE and zero-shot cross-lingual LE transfer) and tasks (binary and graded LE). We report consistent gains over state-of-the-art LE-specialization methods, and successfully LE-specialize word vectors for languages without any external lexical knowledge.

Accepted Paper at Repl4NLP

I am happy to announce that our paper Specializing Distributional Vectors of All Words for Lexical Entailment got accepted to the Repl4NLP workshop at ACL2019 and was declared outstanding paper nominated for the best paper award! We will be giving a 3-minute spotlight oral presentation for the outstanding paper session from 14.45-15.00 on the workshop day, 2 August 2019!