AI News, A Stanford artificial intelligence

Artificial Intelligence That Helps Doctors Predict When Patients Will Die

Advance care planning — which often begins with a simple, structured conversation — can help patients make decisions and settle what will be done ahead of time, relieving some of chaos and confusion that accompanies end-of-life care.

Dr. Stephanie Harman, the clinical chief of palliative care at Stanford Health Care, is leading a pilot program at Stanford Medicine that explores the potential for artificial intelligence (AI) to help doctors guide patients through these decisions.

Though the tool isn't designed to predict a specific time of death — it doesn't give a precise number of months or years — the predictive analytics model identifies patients who have a high probability of dying in three to 12 months.

Would that be something useful, in terms of a decision, a door for clinical care?’ And that was where we said, ‘Wow, yes, that would be useful!’' Taking into account the patient's medical history and millions of other patient records, the AI model uses an algorithm to determine the probability that someone will die within 12 months.

Learning About Machine Learning To assist physicians working with patients nearing the end of their lives, researchers have developed tools for each type of cancer, calibrated for multiple months at a time.

David Hui, a physician in the palliative care department at the University of Texas Anderson Cancer Center in Houston, co-authoreda studythat found a validated prognostic tool called the Palliative Prognostic Index was more accurate than doctors’ estimates when used to determine whether a patient with advanced cancer will diearound 30 days — though notat 100 days.

This clever AI hid data from its creators to cheat at its appointed task

A machine learning agent intended to transform aerial images into street maps and back was found to be cheating by hiding information it would need later in “a nearly imperceptible, high-frequency signal.”

To that end the team was working with what’s called a CycleGAN — a neural network that learns to transform images of type X and Y into one another, as efficiently yet accurately as possible, through a great deal of experimentation.

For instance, skylights on a roof that were eliminated in the process of creating the street map would magically reappear when they asked the agent to do the reverse process: Although it is very difficult to peer into the inner workings of a neural network’s processes, the team could easily audit the data it was generating.

The details of the aerial map are secretly written into the actual visual data of the street map: thousands of tiny changes in color that the human eye wouldn’t notice, but that the computer can easily detect.

street map — all the data needed for reconstructing the aerial photo can be superimposed harmlessly on a completely different street map, as the researchers confirmed: The colorful maps in (c) are a visualization of the slight differences the computer systematically introduced.

In this case the computer’s solution was an interesting one that shed light on a possible weakness of this type of neural network — that the computer, if not explicitly prevented from doing so, will essentially find a way to transmit details to itself in the interest of solving a given problem quickly and easily.



Stanford Artificial Intelligence Laboratory

Statistical Machine Learning Group

Email: adityag at


My research focusses on various aspects of machine learning, including probabilistic modeling, stochastic optimization, and deep learning.

As a Stanford Teaching Fellow, I recently taught a new class on Deep Generative Models in 2018 with an enrollment of 150+ students.

I am interested in developing algorithms for efficient learning and inference in probabilistic models.

A large part of my research in this direction entails the design and analysis of suitable learning objectives, stochastic optimization algorithms, and representation frameworks for probabilistic reasoning (ICLR 2019, AISTATS 2019a, 2018a,b, AAAI 2018a,b).

These endeavors have often led to algorithms that bridge theory and practice for applications across machine learning, e.g., fair representation learning, constraint satisfaction problems, compressed sensing, and multiagent reinforcement learning (AISTATS 2019b, NeurIPS 2018, ICML 2018a,b).

Full Oral Presentation [acceptance rate: 212/2473 (8.6%)] Modeling Sparse Deviations for Compressed Sensing using Generative Models Manik Dhar, Aditya Grover, Stefano Ermon International

Full Oral Presentation [acceptance rate: 212/2473 (8.6%)] Evaluating Generalization in Multiagent Systems using Agent-Interaction Graphs (short) Aditya Grover, Maruan Al-Shedivat, Jayesh K.

Stanford Seminar - Artificial Intelligence: Current and Future Paradigms and Implications

EE380: Computer Systems Colloquium Seminar Artificial Intelligence: Current and Future Paradigms and Implications Speaker: Scott Phoenix, Vicarious ...

Andrew Ng: Artificial Intelligence is the New Electricity

On Wednesday, January 25, 2017, Baidu chief scientist, Coursera co-founder, and Stanford adjunct professor Andrew Ng spoke at the Stanford MSx Future ...

EVERYTHING I learned ABOUT AI at Stanford 2018

In this video I attempt to explain everything I learned about Artificial Intelligence at Stanford 2018. This video is meant for people who wishes to understand the ...

Where AI is today and where it's going. | Richard Socher | TEDxSanFrancisco

Richard Socher is an adjunct professor at the Stanford Computer Science Department where he obtained his PhD working on deep learning with Chris Manning ...

Robo pingpong: Stanford students design, 'teach' robots to play

After learning new software and programming languages, Stanford students in the Artificial Intelligence Laboratory have an opportunity to choose a creative task ...

Artificial Intelligence Applications: AI can Determine Your Sexuality [Stanford Research]

What if Artificial Intelligence, Deep Neural Networks can look at your picture and determine if you are gay or straight? And Could this Technology benefit Fashion ...