Software | AutoML | Technology: Tools, Hardware, and Software

Hyperopt-Sklearn

by

Hyperopt-Sklearn (Hyperparameter optimization for Sklearn) is a Python library for hyperparameter-optimization-based model selection among machine learning algorithms in the Scikit-learn package. The main goal of Hyperopt-Sklearn is to automate and ease the process of hyperparameter tuning for machine learning models. It utilizes Bayesian optimization techniques to decrease the complexity of hyperparameter tuning and speed up the optimization process. It is a valuable tool for tuning hyperparameters and improving performance of Scikit-learn models without manual intervention.

Link: https://hyperopt.github.io/hyperopt-sklearn/

View Resource
AITC Consortium | Basic Information | PennAITech Awardees

Intelligent decision-making tool to connect individuals with AD/ADRD and their caregivers to health app technologies

by

Awardee Organization(s): University of Pittsburgh
Principal Investigator(s): Julie Faieta, PhD, MOT OTR/L
Official Project Title: Health App Review Tool: Connecting those Affected by Alzheimer’s to Needed Technology Support
AITC Partner: PennAITech
Website(s): https://www.shrs.pitt.edu/

The goal of this project is to connect those affected by Alzheimer’s disease and related dementias (ADRD) with effective apps using an intelligent decision-making aid, the Health App Review Tool (HART). The HART is comprised of a User Assessment and App Assessment, that together characterize the features of health apps relative to the needs, abilities, and preferences of individuals with ADRD and their informal caregivers. The HART assesses the goodness of match between user and app variables in order to guide app selection.
The first phase of this pilot project will be used to develop a web-base and app interface to house the HART. The dedicated interface is necessary in preparation for real-world and wide-spread use of the HART. There will be a user interface displaying the HART assessment questions, a back end that completes the scoring process, and a results display. In addition, we will establish a cloud-based library of app scores that can be downloaded and compared to new HART users in the future. The second phase of the project will be a usability study to gather feedback and insight on the HART interfaces for those impacted by ADRD.
The Health App Review Tool (HART) is expected facilitate clinicians, caregivers, and community organizations to select the best apps to meet the unique needs of individuals with ADRD and their caregivers. Improving access to person centered, easy to use technology guidance is intended to increase the impact and equity of app-mediated care.

View Resource
Software | AutoML | Technology: Tools, Hardware, and Software

LAMA: LightAutoML

by

LightAutoML is an open-source Python library aimed at automated machine learning. It is designed to be lightweight and efficient for various tasks with tabular, text data. LightAutoML provides easy-to-use pipeline creation that enables: automatic hyperparameter tuning, data processing; automatic typing, feature selection; automatic time utilization; automatic report creation; and easy-to-use modular scheme to create your own pipelines.

Link: https://lightautoml.readthedocs.io/en/latest/

View Resource
Software | AutoML | Technology: Tools, Hardware, and Software

Ludwig: A low-code framework for building custom AI models like LLMs and other deep neural networks

by

Ludwig is a low-code framework for building custom AI models like LLMs and other deep neural networks. The Ludwig allows you to build custom models with ease. A declarative YAML configuration file is all you need to train a state-of-the-art LLM on your data and its support for multi-task and multi-modality learning. You can also optimize for scale and efficiency, since it also provides automatic batch size selection, distributed training (DDP, DeepSpeed), parameter efficient fine-tuning (PEFT), 4-bit quantization (QLoRA), and larger-than-memory datasets. By supporting hyperparameter optimization, explainability, and rich metric visualizations, you retain full control of your models down to the activation functions. It is modular and extensible and is engineered for production (Docker, HuggingFace).

Link: https://ludwig.ai/latest/

View Resource
Software | AutoML | Technology: Tools, Hardware, and Software

MLBox

by

MLBox is a powerful AutoML Python library that provides fast reading and distributed data preprocessing/cleaning/formatting, highly robust feature selection and leak detection, accurate hyperparameter optimization in high-dimensional space, state-of-the-art predictive models for classification and regression (Deep Learning, Stacking, LightGBM, etc.), and prediction with model interpretation.

Link: https://mlbox.readthedocs.io/en/latest/

View Resource
Software | AutoML | Technology: Tools, Hardware, and Software

MLJAR- supervised: Automated Machine Learning Python package that works with tabular data

by

MLJAR- supervised is an Automated Machine Learning Python package that works with tabular data. It is designed to save time for a data scientist. It abstracts the common way to preprocess the data, construct the machine learning models, and perform hyper-parameters tuning to find the best model. It is no black-box as you can see exactly how the ML pipeline is constructed (with a detailed Markdown report for each ML model). MLJAR- supervised will help you with:
(1) explaining and understanding your data,
(2) trying many different machine learning models,
(3) creating Markdown reports from analysis with details about all models,
(4) saving, re-running and loading the analysis and ML models.

Link: https://supervised.mljar.com/

View Resource
Software | AutoML | Technology: Tools, Hardware, and Software

MLme: Machine Learning Made Easy

by

MLme fulfills the diverse requirements of researchers while eliminating the need for extensive coding efforts by integrating four essential functionalities, namely data exploration, AutoML, CustomML, and visualization. MLme serves as a valuable resource that empowers researchers of all technical levels to leverage ML for insightful data analysis and enhance research outcomes. By simplifying and automating various stages of the ML workflow, it enables researchers to allocate more time to their core research tasks, thereby enhancing efficiency and productivity.

doi: 10.1101/2023.07.04.546825

View Resource
AITC Consortium | Basic Information | PennAITech Awardees

Multimodal conversational AI to assist older adults with daily tasks at home

by

Awardee Organization(s): Pennsylvania State University
Principal Investigator(s): Rui Zhang, PhD | Marie Boltz, PhD, GNP-BC
Official Project Title: Task-Oriented Multimodal Conversational AI for Assisting Seniors with Daily Tasks
AITC Partner: PennAITech
Website(s):
https://www.eecs.psu.edu/
https://ryanzhumich.github.io/

With a global population of over 1 billion people aged 60 and above, there is a rapidly increasing need for innovative age tech solutions to improve the quality of life of older adults. Conversational assistants, powered by cutting-edge technologies in Artificial Intelligence (AI), Natural Language Processing (NLP), and Large Language Models (LLM), are permeating into home care, assisted living, and nursing facilities for smart elderly care. One type of conversational assistant is task- oriented, which can significantly enhance the life experience for senior people by helping them with real-world complex daily tasks. A task-oriented virtual assistant facilitates daily tasks spanning diverse scenarios such as calling for help in response to emergencies, helping with online grocery shopping, recommending cooking recipes, managing smart home devices, and providing financial education and decision-making. It greatly promotes the life quality of older adults by improving their well-being, efficiency, safety, and independence. In this proposal, we design, develop, and deploy a task-oriented multimodal conversational assistant to help older adults with daily tasks. The innovation of this proposal lies in the fact that we will employ a human-centered participatory approach by emphasizing collaboration between designers and end-users through interviewing, prototyping, and testing to address their unique needs and preferences to improve their daily lives.

View Resource
AITC Consortium | Basic Information | PennAITech Awardees

Non-intrusive in-home activities of daily living monitoring using a self-supervised multi-sensor fusion model that detects behavior changes associated with AD/ADRD

by

Awardee Organization(s): University of California San Diego
Principal Investigator(s): Xinyu Zhang, PhD | Alison Moore, MD, MPH
Official Project Title: Non-Intrusive, Fine-Grained In-Home Daily Activity Transcription for Alzheimer’s Monitoring
AITC Partner: PennAITech
Website(s): http://xyzhang.ucsd.edu

Recent research identified a strong correlation between onset of Alzheimer’s disease (AD) and changes in fine physical activities, e.g., movement and dwelling time across locations, daily routines like medicine/water intake. Early detection of such indicators is crucial in compiling better treatment and slowing the progression. Conventional methods for monitoring the activities of daily living (ADL) rely on observation or self-report, which are time consuming, error-prone, and require strict patient compliance. This project aims to transcend such limitations and bridge the key technology gaps in bringing ADL sensing close to clinical practice. The project focuses on the development of EgoADL, a system that uses non-intrusive smartphone/smartwatch sensors to sense ADL. EgoADL builds on a novel self- supervised sensor fusion model that trains itself without user intervention. Instead of classifying among a small known set of ADLs, it directly transcribes raw multi-modal sensor signals into text logs of ADLs which can be interpreted by clinical practitioners or AI models. EgoADL will be the first to use non-visual sensors to transcribe fine ADLs (e.g., human-object interaction) with near-vision precision, in real-time and in a privacy-aware manner. The sensing data can facilitate follow-on clinical and AI analytics, potentially enabling early detection of chronic diseases and safe aging in place. Ubiquitous health monitoring is particularly important for rural and underserved communities, who either do not have access to or cannot afford prolonged hospitalization. EgoADL will be verified through a pilot study in UCSD’s and Upenn’s healthy aging facilities.

View Resource