Josh Susskind @kindsuss.bsky.social · 03/05/2026Apparently there is lots of roundup spraying going on in the forests all over, used as a quick and dirty way to "thin" the growth. Please stop allowing that @governor.ca.gov 000
Josh Susskind @kindsuss.bsky.social · 30/04/2026Their founders are all maga now -- so not surprising. I guess the prospect of having to pay taxes caused them to abandon the whole don't be evil thing. 000
Josh Susskind @kindsuss.bsky.social · 06/04/2026I saw an interview with Terrance Tao talking about this -- that people need to treat deep thinking as a form of workout. 100
Josh Susskind @kindsuss.bsky.social · 07/11/2025Cogsci peeps! This is a great opportunity! @sineadwilliamson.bsky.social is a great mentor and scientist, along with a wonderful team :) 030
Reposted by Josh SusskindMarco Cuturi @marcocuturi.bsky.social · 05/11/2025We have been working with Michal Klein on pushing a module to train *flow matching* models using JAX. This is shipped as part of our new release of the OTT-JAX toolbox (github.com/ott-jax/ott) The tutorial to do so is here: ott-jax.readthedocs.io/tutorials/ne... 1147
Reposted by Josh SusskindMarco Cuturi @marcocuturi.bsky.social · 01/08/2025scaling up the computation of optimal transport couplings to hundreds of thousands of 3k dimensional vectors made easy using sharding and OTT-JAX! check this notebook, it only takes a few lines of code thanks to JAX's native sharding abilities ott-jax.readthedocs.io/en/latest/tu...ott-jax.readthedocs.ioSharded Sinkhorn — ott 0.5.1.dev34+g3462f28 documentation 0142
Josh Susskind @kindsuss.bsky.social · 05/05/2025How about "Machine Learning and Computer Science"? MLCS. 😁 000
Josh Susskind @kindsuss.bsky.social · 23/04/2025What say you @stanford.edu? www.aacu.org/newsroom/a-c...aacu.orgA Call for Constructive Engagement | AAC&UA Call for Constructive Engagement 031
Josh Susskind @kindsuss.bsky.social · 11/04/2025Check out our Apple research work on scaling laws for native multimodal models! Combined with mixtures of experts, native models develop both specialized and multimodal representations! Lots of rich findings and opportunists for follow up research! 065
Josh Susskind @kindsuss.bsky.social · 24/03/2025Paper link: t.co/Z2FZ6YSpbA Code/Model checkpoints: t.co/bXYHZOONOmt.cohttps://arxiv.org/pdf/2502.18435 000
Josh Susskind @kindsuss.bsky.social · 24/03/2025My colleagues in #Apple ML Research posted a fun paper investigating how autoregressive design choices affect reasoning (in this case, multi-choice question answering), showing a benefit to R2L ordering. Reminds me of similar findings for reverse order addition in arxiv.org/abs/2310.16028!arxiv.orgWhat Algorithms can Transformers Learn? A Study in Length GeneralizationLarge language models exhibit surprising emergent generalization properties, yet also struggle on many simple reasoning tasks such as arithmetic and parity. This raises the question of if and when Tra... 193
Josh Susskind @kindsuss.bsky.social · 02/03/2025Permanent Hellcountry is a badass name for a band! Too bad it's also us. stranger than fiction. 000
Josh Susskind @kindsuss.bsky.social · 25/02/2025My colleague Shuangfei Zhai is looking for a summer research intern to work on improving TarFlow at Apple. If interested, send your CV to szhai at apple.com by this week.apple.comAppleDiscover the innovative world of Apple and shop everything iPhone, iPad, Apple Watch, Mac, and Apple TV, plus explore accessories, entertainment, and expert device support. 040
Josh Susskind @kindsuss.bsky.social · 20/02/2025youtu.be/rKBM2kS6B8o Thank you @sanders.senate.gov for speaking upyoutu.beA Sad Moment in American HistoryYouTube video by Senator Bernie Sanders 020
Josh Susskind @kindsuss.bsky.social · 14/02/2025Here's a great paper on scaling laws for teacher-student neural network distillation led by @dbusbridge.bsky.social and Apple colleagues. I've often seen people struggle to get distillation working well enough in practical settings, and I expect the insights in this paper can really help! 040
Josh Susskind @kindsuss.bsky.social · 12/02/2025Here's a fun Apple research paper seeking to understand when/why diffusion models can be composed to generate images containing multiple independent concepts. For example, composing images from a model trained on Preetum's dog and a model trained on hats. Because why wouldn't you want to do that?!! 010
Josh Susskind @kindsuss.bsky.social · 11/02/2025Yeah! that's what I said when I saw it too :) Better than any dog-horse I could make! 020
Josh Susskind @kindsuss.bsky.social · 11/02/2025If you are interested in doing an internship in ML research at Apple, I highly recommend talking with Etai Littwin (and Vimal Thilak is pretty awesome too!) 010
Josh Susskind @kindsuss.bsky.social · 29/01/2025I think it's really important for more of this kind of work to be published openly rather than be walled off due to corporate greed -- scientific inquiry benefits us all. Hopefully we will continue to see lots and lots more of this! 100
Josh Susskind @kindsuss.bsky.social · 29/01/2025This work was born from an Apple internship with Harshay Shah. Samira provided excellent direction and technical contributions along with Vimal, and the entire team was incredibly helpful! I'm intrigued reading comprehension tasks do not follow pre-training scaling curves -- gotta follow this up! 110
Reposted by Josh Susskinddominik1klein.bsky.social @dominik1klein.bsky.social · 23/01/2025Missing the deep learning part? go check out the follow up work @neuripsconf.bsky.social (tinyurl.com/yvf72kzf) and @iclr-conf.bsky.social (tinyurl.com/4vh8vuzk)tinyurl.comGENOT: Entropic (Gromov) Wasserstein Flow Matching with...Single-cell genomics has significantly advanced our understanding of cellular behavior, catalyzing innovations in treatments and precision medicine. However, single-cell sequencing technologies are... 0113
Josh Susskind @kindsuss.bsky.social · 23/01/2025Too disgusted by the Twitter/X vomit and could not justify keeping my account there. Hoping this platform steers clear of disinformation and hate -- and remains a positive place to share science and other good things. 010
Josh Susskind @kindsuss.bsky.social · 23/01/2025Here's a really cool cross-institution study leveraging optimal transport techniques developed by my Apple ML Research colleagues! It's great to see basic research in machine learning translate into scientific tools like this. Cuts into the AI hype a bit ;) 021
Reposted by Josh SusskindCem Koç @cemkoch.bsky.social · 19/12/2024Excited about vision-language models? 🚀 Check out our latest work on FastVLM, a new family of efficient vision-language models that balances the tradeoff between high-resolution image understanding and latency without compromising accuracy! arxiv.org/abs/2412.13303 611
Josh Susskind @kindsuss.bsky.social · 18/12/2024If you're looking for research scientist roles in Europe, check out Marco's post! The Paris team is fantastic, and does diverse idea-driven and impactful research. In addition, MLR is highly collaborative across timezones, so you'd have a chance to work with many others too. 021
Josh Susskind @kindsuss.bsky.social · 10/12/2024Last but not least, please check out the flurry of papers being presented at #NeurIPS2024 , highlighted here in this post machinelearning.apple.com/research/neu... that showcases work from many teams at Apple and their academic collaborators. Thanks for making it to the end ;-)machinelearning.apple.comApple Machine Learning Research at NeurIPS 2024Apple researchers are advancing the field of ML through fundamental research that improves the world’s understanding of this technology and… 020
Josh Susskind @kindsuss.bsky.social · 10/12/2024EC-IJEPA makes the JEPA approach less brittle, and also further unlocks its use in diverse planning and reasoning tasks that leverage pre-tained visual representations as a world model. We're excited to see others build on this work with us! 12/n 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024Returning to the theme of powerful visual representation learning, please check out Vimal Thilak, Etai Littwin, and Anand Gopalakrishnan's EC-IJEPA paper, on improving JEPA models with spatial conditioning: x.com/AggieInCA/st... 11/nx.comx.com 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024WVD is impressive because it enables a range of downstream 3D tasks including 3D view synthesis and depth estimation at inference time by training a good generative model of RGB + XYZ values. Many directions to follow up on here including modeling dynamics! 10/n 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024Moving from images to video and 3D generation, I'm also excited to highlight Jiatao Gu and collaborators' work on WVD (world video diffusion), which jointly models multi-view images and 3D geometry: x.com/thoma_gu/sta... 9/nx.comx.com 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024ASFT is exciting because it shows a single architecture can generate high resolution high quality data and scale well across data domains (we applied it to images and 3D point clouds in this work), removing the need for domain-specific architectural priors. 8/n 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024Continuing in the vein of powerful approaches to image generation, check out our work on ASFT, flow-based Transformers in ambient space, led by Yuyang Wang and Miguel Bautista: twitter.com/YuyangW95/st... 7/ntwitter.com 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024TarFlow demonstrates convincingly that a strong likelihood model can also be a strong generative model, and simplifies the approach to normalizing flows, unlocking the benefits of scale. Try it out and help us extend the approach! 6/n 120
Josh Susskind @kindsuss.bsky.social · 10/12/2024As if AIMv2 wasn't enough, we also opensourced TarFlow, led by Shuangfei Zhai, a Transformer-based normalizing flow model that is both a strong generative model and a state of the art likelihood model: x.com/zhaisf/statu... 5/nx.comx.com 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024Executing on AIMv2 took careful model design and experimentation, smart data filtering for ease of experimentation, and rigorous evaluation. The collaboration generated a lot of excitement as results kept improving. I can’t wait to see how the community builds on this work! 4/n 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024AIMv2 was a herculean effort led by Alaaeldin El-Nouby, Enrico Fini, Mustafa Shukor, and a larger group of Apple collaborators based on the hypothesis that a simple autoregressive approach should enable vision encoders to scale like language models. 3/n 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024First, speaking about opensourcing code and models, I’m really happy to see the community reception around the AIMv2 family of powerful vision encoders (arxiv.org/abs/2411.14402) trained jointly to predict pixels and text. See x.com/alaa_nouby/s... for more details. 2/narxiv.orgMultimodal Autoregressive Pre-training of Large Vision EncodersWe introduce a novel method for pre-training of large-scale vision encoders. Building on recent advancements in autoregressive pre-training of vision models, we extend this framework to a multimodal s... 100
Josh Susskind @kindsuss.bsky.social · 10/12/2024Cross-posting from that other place: I’m really proud of Apple ML research and wanted to share a summary that may be useful for #NeurIPS2024 attendees (and everyone else)! I’m particularly excited about code and model releases and will highlight some here. 1/n 150
Josh Susskind @kindsuss.bsky.social · 10/12/2024Hello Bluesky! I'm a machine learning (aka AI these days) researcher at Apple, and my main interests these days are open science, flatpicking guitar, and rock climbing (and maybe the intersection of all of the above). I'll occasionally highlight some of the research going on in my group here. 040
