[Adapt] [Seminar] Modality Gap in CLIP

[罗嘉鸣]

Hi Adapters!

CLIP is a milestone for multimodal models and contrastive learning strategies. Though it shows surprising ability in zero-shot prediction and much potential for other multimodal tasks, many papers suggest that the alignment of visual and text modalities is not so well as we expected. This mismatch phenomenon is called "modality gap".

In this seminar I will introduce some investigations on the modality gap of CLIP, from the origin to the influence. Wish you would like it!
    
Best regards,
Leo

Time: Wed 10 am. - 11:30 am.
Meeting link: https://teams.microsoft.com/l/meetup-join/19%3ameeting_M2VmMTU5MzgtODUzOC00NmU4LTg0MzktNGFjNDdiMmIwYTI1%40thread.v2/0?context=%7b%22Tid%22%3a%225cdc5b43-d7be-4caa-8173-729e3b0a62d9%22%2c%22Oid%22%3a%221a8b9fa0-af57-4a1c-9390-22d1c201d622%22%7d