Q&A: How to train AI when you don't have enough data
Artificial intelligence excels at sorting through information and detecting patterns or trends. But these machine learning algorithms need to be trained with large amounts of data first.
As researchers explore potential applications for AI, they have found scenarios where AI could be really useful — such as analyzing X-ray image data to look for evidence of rare conditions or detecting a rare fish species caught on a commercial fishing boat — but there's not enough data to accurately train the algorithms.
Jenq-Neng Hwang, University of Washington professor of electrical and computer and engineering, specializes in these issues. For example, Hwang and his team developed a method that teaches AI to monitor how many distinct poses a baby can achieve throughout the day. There are limited training datasets of babies, which meant the researchers had to create a unique pipeline to make their algorithm accurate and useful. The team recently published this work in the IEEE/CVF Winter Conference on Applications of Computer Vision 2024.
UW News spoke with Hwang about the project details and other similarly challenging areas the team is addressing.
Why is it important to develop an algorithm to track baby poses?
Jenq-Neng Hwang: We started a collaboration with the UW School of Medicine and the Korean Electronics and Telecommunications Research Institute's AI Lab. The goal of the project was to try to help families with a history of autism know whether their babies were also likely to have autism. Babies before 9 months don't really have language skills yet, so it's difficult to see if they’re autistic or not. Researchers developed one test, called the Alberta Infant Motor Scale, which categorizes various poses babies can do: If a baby can do this, they get two points; and if they can do that, they get three points; and so on. Then you add up all the points and if the baby is above some threshold, they likely don't have autism.
But to do this test, you need a doctor to observe all the different poses. It becomes a very tedious process because sometimes after three or four hours, we still haven't seen a baby do a specific pose. Maybe the baby could do it, but at that moment they didn't want to. One solution could be to use AI. Parents often have a baby monitor at home. The baby monitor could use AI to continuously and consistently track the various poses a baby does in a day.
Why is AI a good fit for this task?
JNH: My background is studying traditional image processing and computer vision. We were trying to teach computers to be able to figure out human poses from photos or videos, but the trouble is that there are so many variations. For example, even the same person wearing different outfits is a challenging task for traditional image processing to correctly identify that person's elbow on each photo.
But AI makes it so much easier. These models can learn. For example, you could train a machine learning model with a variety of motion captured sequences showing all different kinds of people. These sequences could be annotated with the corresponding 3D poses. Then this model could learn to output a 3D model of a person's pose on a sequence it has never seen before.
But in this case, there aren't a lot of motion captured sequences of babies that also have 3D pose annotations that you could use to train your machine learning model. What did you do instead?
JNH: We don't have a lot of 3D pose annotations of baby videos to train the machine learning model for privacy reasons. It's also difficult to create a dataset where a baby is performing all the possible potential poses that we would need. Our datasets are too small, meaning that a model trained with them would not estimate reliable poses.
But we do have a lot of annotated 3D motion sequences of people in general. So, we developed this pipeline.
First we used the large amount of 3D motion sequences of regular people to train a generic 3D pose generative AI model, which is similar to the model used in ChatGPT and other GPT-4 types of large language models.
We then finetuned our generic model with our very limited dataset of annotated baby motion sequences. The generic model can then adapt to the small dataset and produce high quality results.
Are there other tasks like this: good for AI, but there's not a lot of data to train an algorithm?
JNH: There are many types of scenarios where we don't have enough information to train the model. One example is a rare disease that is diagnosed by X-rays. The disease is so rare that we don't have enough X-ray images from patients with the disease to train a model. But we do have a lot of X-rays from healthy patients. So, we can use generative AI again to generate the corresponding synthetic X-ray image without disease, which can then be compared with the diseased image to identify disease regions for further diagnosis.
Autonomous driving is another example. There are so many real events you cannot create. For example, say you are in the middle of driving and a few leaves blow in front of the car. If you use autonomous driving, the car might think something is wrong and slam on the brakes, because the car has never seen this scenario before. This could result in an accident.
We call these "long-tail" events, which means that they are unlikely to happen. But in daily life we always see random things like this. Until we figure out how to train autonomous driving systems to handle these types of events, autonomous driving cannot be useful. Our team is working on this problem by combining data from a regular camera with radar information. The camera and radar persistently check each other’s decisions, which can help a machine learning algorithm make sense of what's happening.
Additional co-authors on the baby poses paper are Zhuoran Zhou, a UW research assistant in the electrical and computer engineering department; Zhongyu Jiang and Cheng-Yen Yang, UW doctoral students in the electrical and computer engineering department; Wenhao Chai, a UW master's student studying electrical and computer engineering; and Lei Li, a doctoral fellow at the University of Copenhagen. This research was funded by the Electronics and Telecommunications Research Institute of Korea, the National Oceanic and Atmospheric Administration and Cisco Research.
For more information, contact Hwang at hwang@uw.edu.
ARTICLE TITLE
Efficient Domain Adaptation via Generative Prior for 3D Infant Pose Estimation
Generative AI develops potential new drugs for antibiotic-resistant bacteria
Stanford Medicine researchers devise a new artificial intelligence model, SyntheMol, which creates recipes for chemists to synthesize the drugs in the lab.
STANFORD MEDICINE
With nearly 5 million deaths linked to antibiotic resistance globally every year, new ways to combat resistant bacterial strains are urgently needed.
Researchers at Stanford Medicine and McMaster University are tackling this problem with generative artificial intelligence. A new model, dubbed SyntheMol (for synthesizing molecules), created structures and chemical recipes for six novel drugs aimed at killing resistant strains of Acinetobacter baumannii, one of the leading pathogens responsible for antibacterial resistance-related deaths.
The researchers described their model and experimental validation of these new compounds in a study published March 22 in the journal Nature Machine Intelligence.
“There’s a huge public health need to develop new antibiotics quickly,” said James Zou, PhD, an associate professor of biomedical data science and co-senior author on the study. “Our hypothesis was that there are a lot of potential molecules out there that could be effective drugs, but we haven’t made or tested them yet. That’s why we wanted to use AI to design entirely new molecules that have never been seen in nature.”
Before the advent of generative AI, the same type of artificial intelligence technology that underlies large language models like ChatGPT, researchers had taken different computational approaches to antibiotic development. They used algorithms to scroll through existing drug libraries, identifying those compounds most likely to act against a given pathogen. This technique, which sifted through 100 million known compounds, yielded results but just scratched the surface in finding all the chemical compounds that could have antibacterial properties.
“Chemical space is gigantic,” said Kyle Swanson, a Stanford computational science doctoral student and co-lead author on the study. “People have estimated that there are close to 1060 possible drug-like molecules. So, 100 million is nowhere close to covering that entire space.”
Hallucinating for drug development
Generative AI’s tendency to “hallucinate,” or make up responses out of whole cloth, could be a boon when it comes to drug discovery, but previous attempts to generate new drugs with this kind of AI resulted in compounds that would be impossible to make in the real world, Swanson said. The researchers needed to put guardrails around SyntheMol’s activity — namely, to ensure that any molecules the model dreamed up could be synthesized in a lab.
“We’ve approached this problem by trying to bridge that gap between computational work and wet lab validation,” Swanson said.
The model was trained to construct potential drugs using a library of more than 130,000 molecular building blocks and a set of validated chemical reactions. It generated not only the final compound but also the steps it took with those building blocks, giving the researchers a set of recipes to produce the drugs.
The researchers also trained the model on existing data of different chemicals’ antibacterial activity against A. baumannii. With these guidelines and its building block starting set, SyntheMol generated around 25,000 possible antibiotics and the recipes to make them in less than nine hours. To prevent the bacteria from quickly developing resistance to the new compounds, researchers then filtered the generated compounds to only those that were dissimilar from existing compounds.
“Now we have not just entirely new molecules but also explicit instructions for how to make those molecules,” Zou said.
A new chemical space
The researchers chose the 70 compounds with the highest potential to kill the bacterium and worked with the Ukrainian chemical company Enamine to synthesize them. The company was able to efficiently generate 58 of these compounds, six of which killed a resistant strain of A. baumannii when researchers tested them in the lab. These new compounds also showed antibacterial activity against other kinds of infectious bacteria prone to antibiotic resistance, including E. coli, Klebsiella pneumoniae and MRSA.
The scientists were able to further test two of the six compounds for toxicity in mice, as the other four didn’t dissolve in water. The two they tested seemed safe; the next step is to test the drugs in mice infected with A. baumannii to see if they work in a living body, Zou said.
The six compounds are vastly different from each other and from existing antibiotics. The researchers don’t know how their antibacterial properties work at the molecular level, but exploring those details could yield general principles relevant to other antibiotic development.
“This AI is really designing and teaching us about this entirely new part of the chemical space that humans just haven’t explored before,” Zou said.
Zou and Swanson are also refining SyntheMol and broadening its reach. They’re collaborating with other research groups to use the model for drug discovery for heart disease and to create new fluorescent molecules for laboratory research.
The study was funded by the Weston Family Foundation, the David Braley Centre for Antibiotic Discovery, the Canadian Institutes of Health Research, M. and M. Heersink, the Chan-Zuckerberg Biohub, and the Knight-Hennessy scholarship.
JOURNAL
Nature Machine Intelligence
METHOD OF RESEARCH
Computational simulation/modeling
SUBJECT OF RESEARCH
Not applicable
ARTICLE TITLE
Generative AI for designing and validating easily synthesizable and structurally novel antibiotics
No comments:
Post a Comment