Zero-shot prediction with CARPRT method
So, considerando o paper que lemos a respeito do literature review. Really importante essa tarefa. E como importante really is regarding intuition and relations, man. Com esse paper, e discutindo com o Claude, que explicou algo sobre anisotropia, lembrei que li sobre o LeJEPA. Eu não lembrava do nome do paper, sabia que era algum paper do LeCun e que discutia essa questão de anisotropia e isotropia. Discuti com o Claude a respeito, que achou o paper e discutimos, eu propus, actually, em adaptar o LeJEPA no contexto dessa tarefa e do modo training-free, inference time. So, that was great.
Fiquei pensando sobre como establish uma linha de pesquisa. Like, it seems without focus but it not that. The linha de pesquisa seria deep learning resourcefulness, I established now the linha de pesquisa.
There are fundamental problems, like, CARPRT is in the zero-shot prediction with VLMs considering training-free. O LeJEPA is in the self-supervised learning realm.
Man, that is the way, like, really the originality. Se eu ficar buscando establish um field e ficar nisso, just constraint and limitation regarding to think original ideas.
Like, it is not lack of focus, it is interest in solve fundamental problems, the main one regarding the learning itself, intelligence.
So, great, to consider this perspective to use LeJEPA perspective in the zero-shot prediction with VLMs, so that the contribution can be really original considering the geometry proofs in the self-supervised learning perspective for solving a zero-shot prediction training-free problem.
Really great learning.