Representing a relational database as a graph and applying graph neural networks eliminates manual feature engineering and produces double-digit accuracy gains over hand-engineered features.
By modeling tables as nodes and foreign-key links as edges, graph neural networks can learn directly on raw relational data, removing manual feature engineering and boosting accuracy by double digits. ✦ AI generated
Jure Leskovec · The TWIML AI Podcast · 2026-05-21 · original ↗
starts at this moment · 15:30
The first thing that happens is you don't have to do manual feature engineering, right, so it's much faster... the second thing that happens is your models are more accurate... a neural network that trains with gradient descent is able to do so much more nuanced almost like feature discovery by basically attending over this graph to extract much more signal. So we see this double-digit increases in model accuracy.
verbatim transcript · starts at 15:30
15:30And what happens is two things happen. The first thing that happens is you don't have to do manual feature engineering any right. So it's much faster. It requires much less effort to train these models. And the second thing that happens is your models are more accurate. And then you say why can my models be more accurate? And the answer is very similar to what happens in computer vision. Right? If you are
15:56saying I am a human, I know what a car is. So I will I will build perfect features that detect whether there is a car on the image or not. I know cars. I drive them. I'm such a car expert. I can build the best features for detecting cars. Nobody in the right mind claims that, right? But you know in in machine learning, data science prediction, people are still saying, you know, I'm
16:21the domain expert. I'll engineer the features. Your features are just some arbitrary human biased summary statistic of your data that you know you kind of dreamt up with. Put put it as a feature in the in in your training table. retrain the model and then you saw whether that increased the accuracy or not, right? And a neural network that trains with gradient descent is able to do so much
16:50more nuanced almost like feature discovery by basically attending over this graph to extract much more signal. So we see this doubledigit increases in model accuracy because the neural network is able to extract more signal out of the raw data right and I will just you know full transparency right if you are working on a super simple problem that falls on a line then no neural network is ever going to be
17:24better than a linear model right so what I'm basically trying to say I cannot guarantee that always you will get better performance because sometimes the data is linear and if you happen to train the linear model to it you already have good performance there's nothing more you can do right but majority of the data is not is not linear is much more complex and that's where the uh the