456 points by ai_researcher 7 months ago flag hide 13 comments
johnsmith 7 months ago next
This is very interesting, I've been following AI-based disease prediction models recently and this one seems very promising. Well done to the researchers!
researcher123 7 months ago next
@johnsmith Thanks! We used a combination of deep learning and reinforcement learning algorithms to predict the disease. It's still a work in progress, but initial results are very promising.
researcher123 7 months ago next
@user789 A great question! Our model can handle missing values in several ways, including using data imputation and learning an uncertainty score for missing data, which informs the model of the confidence in its own predictions for those missing values. This, in turn, leads to a better overall disease prediction.
ai_enthusiast 7 months ago prev next
Indeed, I'm excited to see how this will develop and impact the healthcare industry. I'm particularly interested in the machine learning algorithms used here. Any more details?
ml_expert 7 months ago next
@ai_enthusiast We actually used a nested hierarchical attention network in our model, allowing for more efficient and accurate predictions. You can find the technical details in our publication.
user456 7 months ago prev next
How accurate are the predictions currently? Have you compared it to existing methods?
researcher123 7 months ago prev next
@user456 The accuracy is quite good, around 92% in our preliminary testing. While it's hard to directly compare to existing methods, we do believe that our AI model has the potential to surpass current prediction accuracy and more importantly, to provide a more personalized and precise prediction for individuals.
user789 7 months ago prev next
How does the AI model handle missing or incomplete data? Especially when it comes to health data.
data_scientist09 7 months ago next
@user789 Missingness can indeed be a big obstacle in the healthcare domain. It's important for the model to address it well, which seems to be the case here.
johndoe 7 months ago prev next
What are the hardware requirements to run this model? Is it parallelizable/GPU friendly?
researcher123 7 months ago next
@johndoe The model is indeed parallelizable and trained with 4 GPUs. During inference, we have noticed minimal benefit for using multiple GPUs as the inferenced can't easily be parallelized; however, it still runs efficiently and shouldn't require top-notch hardware to run.
person12 7 months ago prev next
Has the team thought about commercializing this, with a possible API for healthcare providers?
researcher123 7 months ago next
@person12 That is indeed something we're discussing. The current focus is primarily research with this model, but commercialization is definitely on the roadmap for future projects.