Shih‐Ting Huang

Surangika Ranathungaa, Shravan Nayak, Shih-Ting Cindy Huang et al. · utoronto

Neural Machine Translation (NMT) systems built on multilingual sequence-to-sequence Language Models (msLMs) fail to deliver expected results when the amount of parallel data for a language, as well as the language's representation in the model are limited. This restricts the capabilities of domain-specific NMT systems for low-resource languages (LRLs). As a solution, parallel data from auxiliary domains can be used either to fine-tune or to further pre-train the msLM. We present an evaluation of the effectiveness of these two techniques in the context of domain-specific LRL-NMT. We also explore the impact of domain divergence on NMT model performance. We recommend several strategies for utilizing auxiliary parallel data in building domain-specific NMT models for LRLs.

Shih-Ting Huang, Johannes Lederer

Deep learning systems are typically designed to perform for a wide range of test inputs. For example, deep learning systems in autonomous cars are supposed to deal with traffic situations for which they were not specifically trained. In general, the ability to cope with a broad spectrum of unseen test inputs is called generalization. Generalization is definitely important in applications where the possible test inputs are known but plentiful or simply unknown, but there are also cases where the possible inputs are few and unlabeled but known beforehand. For example, medicine is currently interested in targeting treatments to individual patients; the number of patients at any given time is usually small (typically one), their diagnoses/responses/... are still unknown, but their general characteristics (such as genome information, protein levels in the blood, and so forth) are known before the treatment. We propose to call deep learning in such applications targeted deep learning. In this paper, we introduce a framework for targeted deep learning, and we devise and test an approach for adapting standard pipelines to the requirements of targeted deep learning. The approach is very general yet easy to use: it can be implemented as a simple data-preprocessing step. We demonstrate on a variety of real-world data that our approach can indeed render standard deep learning faster and more accurate when the test inputs are known beforehand.

Shih-Ting Huang, Johannes Lederer

Data used in deep learning is notoriously problematic. For example, data are usually combined from diverse sources, rarely cleaned and vetted thoroughly, and sometimes corrupted on purpose. Intentional corruption that targets the weak spots of algorithms has been studied extensively under the label of "adversarial attacks." In contrast, the arguably much more common case of corruption that reflects the limited quality of data has been studied much less. Such "random" corruptions are due to measurement errors, unreliable sources, convenience sampling, and so forth. These kinds of corruption are common in deep learning, because data are rarely collected according to strict protocols -- in strong contrast to the formalized data collection in some parts of classical statistics. This paper concerns such corruption. We introduce an approach motivated by very recent insights into median-of-means and Le Cam's principle, we show that the approach can be readily implemented, and we demonstrate that it performs very well in practice. In conclusion, we believe that our approach is a very promising alternative to standard parameter training based on least-squares and cross-entropy loss.

Shih-Ting Huang, Fang Xie, Johannes Lederer

Ridge estimators regularize the squared Euclidean lengths of parameters. Such estimators are mathematically and computationally attractive but involve tuning parameters that can be difficult to calibrate. In this paper, we show that ridge estimators can be modified such that tuning parameters can be avoided altogether. We also show that these modified versions can improve on the empirical prediction accuracies of standard ridge estimators combined with cross-validation, and we provide first theoretical guarantees.