• ashwinvis@mastodon.acc.sunet.seOP
    link
    fedilink
    arrow-up
    1
    ·
    2 months ago

    @askphysics also this is why we need regularisation for #MachineLearning models, isn’t it? The weights are the coordinates in a hyper space, the model resides in. Without regularization, we risk the weights along only one or two coordinates being dominant and the other coordinates don’t get any “volume”.