New paper at #ICLR2022! Image classifiers generalize poorly into new classes without forgetting the old ones (catastrophic forgetting), especially when there are few examples from novel classes (few-shot). People say fine-tuning is the worst, but we disagree! 😮⬇️

#PrincetonU will increase graduate fellowship and stipend rates by an average of 25% to about $40,000 for doctoral candidates during the 10-month academic year. It is the University’s largest one-year increase in graduate student stipend rates. bit.ly/3IzeArZ

Are you a PhD student interested in machine learning and numerical modeling for weather & climate? My team at Google Research is looking to hire a student researcher for this summer (and likely beyond) in either Mountain View, CA or Cambridge, MA.

Machine Learning lives in an uncanny valley btw Science and Engineering. It's the worst of both worlds. We don't care about understanding, just making things "work" (bad science). We don't care if things work in the real world, just on contrived benchmarks (bad engineering).

Deep learning is cool because it leads to double bugs. Is it a bug in your code, is it a bug in your idea? 🤷‍♂️

🔥The WILDS 2.0 benchmark is an oral at ICLR'22! Existing distribution shift benchmarks do not reflect the breadth of scenarios that arise in real-world applications. Wilds 2.0 extends the Wilds benchmark to include curated unlabeled data for DA research. arxiv.org/abs/2112.05090

Someone at Pixar deleted all of Toy Story 2 and the backup hadn't worked for a month, and the only reason we saw that movie was b/c someone on maternity leave had a copy of it on her home computer. Her name is Galyn Susman and she is now the producer for the new Lightyear movie!

Dear people who review papers. The following are generally unacceptable to put in a review: 1) Too simple 2) Not novel

Hello world! Welcome to the official account of the ViSDA NeurIPS Conference challenge! This year, we focus on sim2real domain adaptive semantic segmentation for automated waste sorting. $3000 and invitations to present to top-3 teams. More at ai.bu.edu/visda-2022/!

This is shaping up to be an awesome competition! Training data released next week.

Evaluation server and interface is up on eval.ai. Find submission instructions at github.com/dbash/visda202…!

Need something to do this weekend? Get a head start on our ViSDA 2022 challenge. $3000 in prizes and a chance to present at #NeurIPS2022! This year, we’re tackling waste sorting.Recycling is imperfect and many recyclables get placed in the landfill. [1/3]

Care about computer vision and the environment? Join our computer vision for recycling challenge ⬇️

Which one do you prefer when *reading* a paper? Please, share to increase the sample size.

Just under a month until the testing stage of VISDA 2022. There’s still time to submit! Help push vision-based recycling sorting forward, get up to a $2000 prize, and get a chance to present at our workshop. More details in pinned tweet and at ai.bu.edu/visda-2022/.

MAGE: MAsked Generative Encoder to Unify Representation Learning and Image Synthesis abs: arxiv.org/abs/2211.09117

Harmful Algal Blooms are serious environmental issue, but predictive modeling can help mitigate! So cool to see initiatives like this! And prizes aren't bad either...consider fighting holiday boredom with some hacking for this challenge!

Txt2img just keeps getting better and better...and MUSE can even kind of count. Amazing.

COLA: How to adapt vision-language models to Compose Objects Localized with Attributes? abs: arxiv.org/abs/2305.03689 paper pages: huggingface.co/papers/2305.03…