Visual recognition in low-data regimes requires deep neural networks to learn generalized representations from limited training samples. Recently, CLIP-based methods have shown promising few-shot performance benefited from the contrastive language-image pre-training. We then question, if the more diverse pre-training knowledge can be cascaded to further assist few-shot representation learning. In this paper, we propose CaFo, a Cascade of Foundation models that incorporates diverse prior knowledge of various pre-training paradigms for better few-shot learning. Our CaFo incorporates CLIP's language-contrastive knowledge, DINO's vision-contrastive knowledge, DALL-E's vision-generative knowledge, and GPT-3's language-generative knowledge. Specifically, CaFo works by 'Prompt, Generate, then Cache'. Firstly, we leverage GPT-3 to produce textual inputs for prompting CLIP with rich downstream linguistic semantics. Then, we generate synthetic images via DALL-E to expand the few-shot training data without any manpower. At last, we introduce a learnable cache model to adaptively blend the predictions from CLIP and DINO. By such collaboration, CaFo can fully unleash the potential of different pre-training methods and unify them to perform state-of-the-art for few-shot classification. Code is available at https://github.com/ZrrSkywalker/CaFo.

But Jenna Lester, a dermatologist at the University of California San Francisco, was growing frustrated with the poor quality images she'd receive of her dark-skinned patients. It wasn't just a cosmetic issue -- the bad photos meant darker-skinned people weren't getting the same quality of care. So in January, Lester co-authored a paper in the British Journal of Dermatology that gives a step-by-step guide to photographing skin of color accurately in clinical settings. Lester, who herself is Black, said, "I feel like these issues and my life is constantly me saying, 'Hey, what about us?' 'What about these patients?'" Medical photographs are vital to documenting disease in textbooks and journals and training medical students.

Samsung's new high-end smartphone alerts owners to problems with their photos as soon as they are snapped. The Galaxy Note 9 uses software to identify if the images are blurry or whether the subject is blinking, among other issues. The innovation is one of several artificial intelligence-enhanced abilities the handset offers that do not rely on remote computer servers. Analysts say this should help tackle increasing competition from Huawei. The Chinese rival made much of its own flagship phone's machine learning abilities when it launched in March. The Note 9's other new features include a stylus that wirelessly communicates with the phone to act as a remote control.

A team of researchers from NVIDIA, MIT and Aalto University have found a way to fix pixelated photographs using AI -- even if the AI has never seen a clean example of the target photo. The group used deep learning -- a type of machine learning that can teach AI to piece together images, text or video -- to restore images with noisy input. While previous work trained AI to reconstruct photos with missing facial features by showing it complete photos, the current method means AI can rebuild a clean photo by only using "corrupted data", or two tarnished images. And surprisingly, its ability to clean up artifacts, remove text and beautify photos occasionally produced a better outcome than methods requiring cleaner reference material. The AI does this by utilizing a neural network that's been trained using corrupt photos.

How did Google get Clips, its AI-powered camera, to learn to automatically take the best shots of users and their families? Well, as the company explains in a new blog post, its engineers went to the professionals -- hiring "a documentary filmmaker, a photojournalist, and a fine arts photographer" to produce visual data to train the neural network powering the camera. The blog post explains this process in a little more detail, but it's basically what you'd expect for this sort of AI. In order for the software to recognize what makes a good or a bad photo, it had to be fed lots of examples. The programmers thought about not only obvious markers (eg, it's a bad photo if there is blurring or if something's covering the lens) but also more abstract criteria, such as "time" -- training Clips with the rule, "Don't go too long without capturing something."