principal machine learning engineer working mostly on computer vision
Seattle
Joined June 2009
- Tweets1.9K
- Following272
- Followers235
- Likes2.2K
cesar romero retweeted
Introducing MusicGen Web: AI-powered music generation directly in your browser, built with 🤗 Transformers.js! 🎵
Everything runs 100% locally, meaning no calls to an API! 🤯 Served as a static website... this costs $0 to host and run! 🔥
Try it out yourself! 👇
cesar romero retweeted
Google just launched VideoPoet, and it's insane
It's literally ChatGPT for:
- Text-to-Video
- Image-to-Video
- Inpainting
- Video Editing
Here're 10 wild examples:
cesar romero retweeted
Conway’s Game of Life is Omniperiodic
arxiv.org/abs/2312.02799
“In the theory of cellular automata, an oscillator is a pattern that repeats itself after a fixed number of generations; that number is called its period. A cellular automaton is called omniperiodic if there exist oscillators of all periods. At the turn of the millennium, only twelve oscillator periods remained to be found in Conway's Game of Life. The search has finally ended, with the discovery of oscillators having the final two periods, 19 and 41, proving that Life is omniperiodic. Besides filling in the missing periods, we give a detailed history of the omniperiodicity problem and the strategies used to solve it, summarising the work of a large number of people in the decades since the creation of Life.”
cesar romero retweeted
Gemini is a line of 3 AI models -- the largest one (Ultra) scores better than GPT-4V across a wide range of text & multimodal benchmarks. It is also the first model to outperform human experts on MMLU.
The Pro version (smaller) is already available in Bard. Gemini Ultra will be available as an API early next year for developers and the enterprise market.
Announcement: deepmind.google/technologies…
Technical report: storage.googleapis.com/deepm…
cesar romero retweeted
Excited to announce Imagen Video, our new text-conditioned video diffusion model that generates 1280x768 24fps HD videos! #ImagenVideo
imagen.research.google/video…
Work w/ @wchan212 @Chitwan_Saharia @jaywhang_ @RuiqiGao @agritsenko @dpkingma @poolio @mo_norouzi @fleet_dj @TimSalimans
cesar romero retweeted
imo #stablediffusion release today is a day of historic proportion for human creativity, with so much human visual creativity bottled up into one accessible artifact. Big part of a phase shift into an era of human+AI art collab that we’ve just barely scratched the surface of.
Delighted to announce the public open source release of #StableDiffusion!
Please see our release post and retweet! stability.ai/blog/stable-dif…
Proud of everyone involved in releasing this tech that is the first of a series of models to activate the creative potential of humanity
cesar romero retweeted
Breaking News: Stanford University will receive a record $1.1 billion gift from the venture capitalist John Doerr to fund a school focused on climate change and sustainability. nyti.ms/39vIAbX
cesar romero retweeted
I can't be the only one noticing that Putin's response to failure in Ukraine is extremely Trumpian. 1. Deny that anything has gone wrong 2. Insist that your own decisions have always been perfect 3. Whine about cancel culture
Tough guys, both
cesar romero retweeted
ArXiv arxiv.org/abs/2203.00854: Speeding up AlphaFold: reduces training time from 11 days to 67 hours and 7.5∼9.5× speedup for long sequences. GPU optimizations. With github repository.
cesar romero retweeted
I realized this morning that the same number of years have now passed since the Transformer paper came out as between the AlexNet paper and the Transformer paper. Definitely makes me feel my age a bit...
cesar romero retweeted
Perceiver. Being able to treat every modality as a sequence of bytes has been a personal deep learning dream. Perceiver is a transformer-derived architecture proposing a few modifications to achieve this.
Papers: arxiv.org/abs/2103.03206 & arxiv.org/abs/2107.14795
cesar romero retweeted
As a…
– Crypto bro
I want to…
– tell everyone we are taking back control from the evil wealth-driven ultra-rich
so that…
– I too may become ultra-rich
ALT Image repeats text of tweet content in image form for some reason
cesar romero retweeted
In case you missed it - the 2021 DeepMind x @ai_ucl Reinforcement Learning Lecture Series is available online.
Catch up and watch the entire series via dpmd.ai/2021RLseries
Introducing the '21 DeepMind x @ai_ucl Reinforcement Learning Lecture Series, a comprehensive introduction to modern RL.
Follow along with our researchers are they explore Markov Decision Processes, sample-based learning algorithms & much more: dpmd.ai/2021RLseries 1/2
cesar romero retweeted
Speeding Up Reinforcement Learning with a New Physics Simulation Engine
Blog post by @bucketofkets on Brax, a new physics simulation engine that matches the performance of a large compute cluster with just a single TPU or GPU.
ai.googleblog.com/2021/07/sp…
cesar romero retweeted
Sharing "The Benchmark Lottery" from @GoogleAI & @DeepMind.
In this meta-paper (arxiv.org/abs/2107.07002), we examine the challenges of ML benchmarking (e.g., model comparisons) and how it affects long-term progress. 1/
cesar romero retweeted
TorchShard is a lightweight engine for slicing a PyTorch tensor into parallel shards. It can reduce GPU memory and scale up the training when the model has massive linear layers or huge classes. Read more below:
medium.com/pytorch/torchshar…
cesar romero retweeted
How neat would it be if your game’s AI could learn to collaborate on its own? What if we told you... it already can!
Presenting the DodgeBall environment, our latest #MLAgents demo with cooperative agents trained through reinforcement learning 🧠 ⬇️ on.unity.com/3kfSbap