IBM’s goal is to replace the best guesses with database decisions through a new causal inference toolkit.

Researchers are using this analysis to discover new uses for existing drugs and determine the impact of COVID-19 on medical screening.


Everyone has a premonition of what keeps them healthy or makes them sick, whether or not there is evidence to support those theories. This is a complex calculation because multiple factors, such as genetics, living conditions, environmental factors, diet, family history, exercise, and financial status, interact to affect human health. IBM’s new causal inference toolkit aims to analyze multiple factors in similarly complex situations to determine what makes a difference and what doesn’t. The idea is to replace the best guesses with data-backed decisions.

Causal inference is an analysis method that follows: Consider assumptions, study design and estimation strategies This allows researchers to draw causal conclusions based on the data. IBM’s goal for this website is to allow data scientists to quantify the causality of their data. The tutorials on this site are:

  • How Does Smoking Cessation Affect Weight Loss?
  • Do Agricultural Technology Affect Water Pollution?
  • How do marketing campaigns affect long-term bank deposits and purchases?
  • Does vocational training increase the income of disadvantaged people?

NS Causal Inference 360 ​​Open Source Toolkit Includes tutorials, background information, and demonstrations. This analysis is relevant in many sectors, including healthcare, agriculture, finance and banking marketing.

IBM has built an open source site to bring “a long-standing machine learning technique to the field of causal inference.” This resource includes how to train a causal model, how to evaluate it to choose the most appropriate method, the underlying model, and parameter adjustment tactics.

The company uses a toolkit Research new uses for existing prescription drugs At a laboratory in Haifa, Israel. In a blog post, IBM researcher Michal Rosen-Zvi explained that the team found that drugs used to treat insomnia could treat dementia, which often develops in Parkinson’s disease.

Researchers created a virtual clinical trial of simulated patients and evaluated their effectiveness according to the results recorded in electronic health records and claims. The team looked for drugs that had a statistically significant effect on both EHR and claim data to consider diversion. As Rosen-Zvi explained, “analysis reveals the therapeutic benefits of two drugs that reduce the incidence of population-level dementia associated with Parkinson’s disease.”

“Our research is an important use case, but it’s very likely that other drugs will be reused for a variety of neurodegenerative diseases and infections. AI can be very helpful,” she said. I concluded.

Rosen-Zvi was also one of the researchers Analyzed medical data to understand why women skipped breast cancer screening appointments in 2020.. The team applied advanced machine learning techniques to known predictors of this common problem in healthcare and new factors that could affect behavior. After considering confounding bias, the team used causal inference techniques to infer the impact of closure on no-show. In a pre-published research paper, researchers state that the results “suggest that patient-perceived cancer risk and time-based factors for COVID-19 are key predictors.” increase.

As part of the toolkit release, IBM Open source python library this week. The new features are:

  • New model: matching (estimator and pretreatment transformer); overlap weights; hem
  • The weight model now includes the same fit () API as the result model
  • Updated Dependencies: Removed Sea Born. 0.25 panda; learn with scikit-0.25

This latest toolkit is part of a collection of open source AI tools that IBM has released over the years to build reliable AI. AI Fairness 360, AI explanation 360, Hostile Robustness 360, AI Fact Sheet 360 When Uncertainty quantification 360..

See also IBM’s goal is to replace the best guesses with database decisions through a new causal inference toolkit.

Back to top button