- This event has passed.
Erlangen AI Hub Seminar: Estimating Intrinsic Dimensionality with L2N2
April 16 @ 1:00 pm - 2:00 pm
Estimating the intrinsic dimensionality (ID) of data is a fundamental problem in machine learning and computer vision, as it reveals the true degrees of freedom underlying high-dimensional observations. In this talk, Eng-Jon Ong (QMUL) introduces L2N2, a simple yet powerful ID estimator based on nearest-neighbour distance ratios that achieves state-of-the-art performance with minimal computational overhead.
We present a theoretical framework showing that L2N2 is universal, in the sense that it provably converges to the true intrinsic dimensionality independently of the underlying data distribution. We complement this analysis with extensive experiments on synthetic benchmark manifolds, demonstrating strong empirical performance.
Beyond controlled settings, we show how L2N2 can be applied to real-world data, including estimating the intrinsic dimensionality of datasets such as MNIST via autoencoder representations. Finally, we illustrate how L2N2 provides new insights into deep learning by tracking how the intrinsic dimensionality of neural network layer representations evolves during training, particularly on the phenomenon of grokking.
Register here: Erlangen AI Hub Seminar: Estimating Intrinsic Dimensionality with L2N2 (Eng-Jon Ong)