About
The Machine Learning for Health and Well-Being (MLwell) Lab is a research lab at the Bio-Medical Engineering department at Tel-Aviv University. Our vision is to create the technology to allow everyone and everywhere access to personalized medicine and precision psychology that is: (i) effective (ii) respects the biological, cultural and behavioral differences between people (iii) respects privacy and other ethical requirements (iv) affordable. Our mission is to improve the state in the art in machine learning algorithms for personalized medicine and precision psychology.
Our News
10.11.24
Upcoming Event
Context-Aware Automated Quality Evaluation of Structured Health Records" will be presented at IDSAI2025 on January 7th, 2025
In this work, we address the challenge of ensuring data quality in Electronic Health Records (EHRs), where the focus often shifts to model development rather than the underlying data itself. To fill this gap, we introduce the Medical Data Pecking Tool (MDPT), an innovative solution that utilizes unit-testing techniques to evaluate EHR data quality and its suitability for specific research questions. By combining a dataframe testing tool with a Large Language Model (LLM), MDPT can automatically generate and execute customized evaluations based on predefined criteria such as population traits and regional health patterns, ensuring that the data aligns with expected patterns for various diseases and geographic regions.
10.11.24
New Paper
The Intelligible and Effective Graph Neural Additive Networks
The Graph Neural Additive Network (GNAN) is the first interpretable-by-design graph neural network. It extends Generalized Additive Models (GAMs) to graph data, offering both high performance and transparency. As a result, GNAN is well-suited for high-stakes applications
10.11.24
New Paper
Lost in Translation: The Limits of Explainability in AI
This paper examines whether eXplainable AI (XAI) tools can effectively support the legal "right to explanation" by analyzing explanation's role across different stakeholders - decision subjects, decision makers, and the broader ecosystem. While XAI proves effective in strengthening system authority from an ecosystem perspective, it falls short in serving both decision subjects' and makers' needs, potentially making it an inadequate and possibly harmful tool for protecting human rights rather than the guardian it was intended to be.
12.9.24
xgbGAMView
Generalized Additive Models (GAMs) based on xgboost with smoothing and scikit-learn compatible interface
The xgbGAMView allows learning GAMs with a familiar scikit-learn interface. The GAMs use xgboost as the underlying engine to learn the model and offer visualization as well as graph smoothing options. The library can be installed from PyPi (pip install xgbGAMView).
15.4.24
New Paper
Impact of Long-COVID in children: a large cohort study
The impact of long-term Coronavirus disease 2019 (COVID-19) on the pediatric population is still not well understood. This study was designed to estimate the magnitude of COVID-19 long-term morbidity 3–6 months after the date of diagnosis.
24.3.24
New Paper
TREE-G: Decision Trees Contesting Graph Neural Networks
When dealing with tabular data, models based on decision trees are a popular choice due to their high accuracy on these data types, their ease of application, and explainability properties. However, when it comes to graph-structured data, it is not clear how to apply them effectively, in a way that in- corporates the topological information with the tabular data available on the vertices of the graph. To address this challenge, we introduce TREE-G. TREE-G modifies standard decision trees, by introducing a novel split function that is specialized for graph data. Not only does this split function incorporate the node features and the topological information, but it also uses a novel pointer mechanism that allows split nodes to use information computed in previous splits. Therefore, the split function adapts to the predictive task and the graph at hand. We analyze the theoretical properties of TREE-G and demonstrate its benefits empirically on multiple graph and vertex prediction benchmarks. In these experiments, TREE-G consistently outperforms other tree-based models and often outperforms other graph-learning algorithms such as Graph Neural Networks (GNNs) and Graph Kernels, sometimes by large margins. Moreover, TREE-Gs models and their predic tions can be explained and visualized.
14.3.24
Announcement
Congratulations to Belle Kriger, Hagar Rosenblatt, Chaya Ben-Yehuda, and Yarin Udi for completing their Master's degrees
Congratulations to Belle Kriger, Hagar Rosenblatt, Chaya Ben-Yehuda, and Yarin Udi for completing their Master's degrees
16.2.24
Announcement
Tree-G: a new tool for learning over graphs
Graphs, such is protein interaction graphs, social contact graphs, and molecular structure graphs, are commonly used. Learning over graphs is challenging and is done using Graph Neural Networks (GNNs) in most cases. We introduce a new method that uses trees and gradient boosting for the task. It outperforms GNNs in many cases. Try it yourself
9.11.23
New Paper
A Work in Progress: Tighter Bounds on the Information Bottleneck for Deep Learning
The field of Deep Neural Nets (DNNs) is still evolving and new architectures are emerging to better extract information from available data. The Information Bottleneck, IB, offers an optimal information theoretic framework for data modeling. However, IB is intractable in most settings. In recent years attempts were made to combine deep learning with IB both for optimization and to explain the inner workings of deep neural nets. VAE inspired variational approximations such as VIB became a popular method to approximate bounds on the required mutual information computations. This work continues this direction by introducing a new tractable variational upper bound for the IB functional which is empirically tighter than previous bounds. When used as an objective function it enhances the performance of previous IB-inspired DNNs in terms of test accuracy and robustness to adversarial attacks across several challenging tasks. Furthermore, the utilization of information theoretic tools allows us to analyze experiments and confirm theoretical predictions in real world problems.
20.9.23
New Paper
Fake, deep-fake, the video law and the principle of narrow interpretation: what is the law?
How should courts act upon deep-fakes? should legislators modify existing laws to address it? In a new paper (in Hebrew) we study these issues and suggest that adding clauses to existing legislation specific to deep-fake might create problems in laws that do not add such clauses due to the principle of narrow interpretation.
20.6.23
Announcement
Congratulations to Daniel, Neta, Lotan, Yuval and Yuval on their great undergrad projects.
Congratulations to Daniel Sarusi, Neta Biran, Lotan Hacohen, Yuval Reingold, and Yuval Argoetti on very successful presentations of their undergrad projects. Click the logo ← to see a short video (in Hebrew) about Yuval A.'s project.
25.5.23
New Paper
The Case Against Explainability
Explainability has been proposed as a possible solution to some of the risks emerging from recent advances in AI. In this paper, we study explanations from legal point of view and show that many of the reasons for requiring explanations cannot be fulfilled by AI systems. Moreover, in some cases, these explanations can increase risks instead of mitigating them.