Micah J. Smith

Eric Lehman, Evan Hernandez, Diwakar Mahajan et al. · mit

Although recent advances in scaling large language models (LLMs) have resulted in improvements on many NLP tasks, it remains unclear whether these models trained primarily with general web text are the right tool in highly specialized, safety critical domains such as clinical text. Recent results have suggested that LLMs encode a surprising amount of medical knowledge. This raises an important question regarding the utility of smaller domain-specific language models. With the success of general-domain LLMs, is there still a need for specialized clinical models? To investigate this question, we conduct an extensive empirical analysis of 12 language models, ranging from 220M to 175B parameters, measuring their performance on 3 different clinical tasks that test their ability to parse and reason over electronic health records. As part of our experiments, we train T5-Base and T5-Large models from scratch on clinical notes from MIMIC III and IV to directly investigate the efficiency of clinical tokens. We show that relatively small specialized clinical models substantially outperform all in-context learning approaches, even when finetuned on limited annotated data. Further, we find that pretraining on clinical tokens allows for smaller, more parameter-efficient models that either match or outperform much larger language models trained on general text. We release the code and the models used under the PhysioNet Credentialed Health Data license and data use agreement.

Micah J. Smith, Jürgen Cito, Kelvin Lu et al.

While the open-source software development model has led to successful large-scale collaborations in building software systems, data science projects are frequently developed by individuals or small teams. We describe challenges to scaling data science collaborations and present a conceptual framework and ML programming model to address them. We instantiate these ideas in Ballet, a lightweight framework for collaborative, open-source data science through a focus on feature engineering, and an accompanying cloud-based development environment. Using our framework, collaborators incrementally propose feature definitions to a repository which are each subjected to an ML performance evaluation and can be automatically merged into an executable feature engineering pipeline. We leverage Ballet to conduct a case study analysis of an income prediction problem with 27 collaborators, and discuss implications for future designers of collaborative projects.

Micah J. Smith, Jürgen Cito, Kalyan Veeramachaneni

Developers in data science and other domains frequently use computational notebooks to create exploratory analyses and prototype models. However, they often struggle to incorporate existing software engineering tooling into these notebook-based workflows, leading to fragile development processes. We introduce Assemblé, a new development environment for collaborative data science projects, in which promising code fragments of data science pipelines can be contributed as pull requests to an upstream repository entirely from within JupyterLab, abstracting away low-level version control tool usage. We describe the design and implementation of Assemblé and report on a user study of 23 data scientists.

Shubhra Kanti Karmaker Santu, Md. Mahadi Hassan, Micah J. Smith et al.

As big data becomes ubiquitous across domains, and more and more stakeholders aspire to make the most of their data, demand for machine learning tools has spurred researchers to explore the possibilities of automated machine learning (AutoML). AutoML tools aim to make machine learning accessible for non-machine learning experts (domain experts), to improve the efficiency of machine learning, and to accelerate machine learning research. But although automation and efficiency are among AutoML's main selling points, the process still requires human involvement at a number of vital steps, including understanding the attributes of domain-specific data, defining prediction problems, creating a suitable training data set, and selecting a promising machine learning technique. These steps often require a prolonged back-and-forth that makes this process inefficient for domain experts and data scientists alike, and keeps so-called AutoML systems from being truly automatic. In this review article, we introduce a new classification system for AutoML systems, using a seven-tiered schematic to distinguish these systems based on their level of autonomy. We begin by describing what an end-to-end machine learning pipeline actually looks like, and which subtasks of the machine learning pipeline have been automated so far. We highlight those subtasks which are still done manually - generally by a data scientist - and explain how this limits domain experts' access to machine learning. Next, we introduce our novel level-based taxonomy for AutoML systems and define each level according to the scope of automation support provided. Finally, we lay out a roadmap for the future, pinpointing the research required to further automate the end-to-end machine learning pipeline and discussing important challenges that stand in the way of this ambitious goal.