Efthymios Tzinis

Senior Research Scientist @GoogleAI | Ph.D. from @IllinoisCS | Formerly @merl_news, @RealityLabs | My opinions do not represent my employer

[Efthymios Tzinis @ETzinis]

My Ph.D. dissertation on "Unsupervised sound separation" is online https://t.co/Fg3d1QG2WC 😎 If you are interested in self-supervised, multi-modal, efficient and federated learning aspects for sound separation, feel free to take a look 👀

[Pedro Domingos @pmddomingos]

Resist the delusional/dishonest attempt by OpenAI and Anthropic to redefine AGI as what they've already done.

[Milton Friedman Daily @MiltonFriedmanW]

“A major source of objection to a free economy is precisely that it does this task so well: it gives people what they want instead of what a particular group thinks they ought to want.”

— Milton Friedman

[Yuchen Jin @Yuchenj_UW]

Larry Page & Sergey Brin had the PageRank paper (the algorithm behind Google Search) rejected. A reviewer called it “disjointed.”

Geoffrey Hinton's Dropout was rejected for being “too simple.”

I often feel the academic peer review is like a random process, especially when a paper is very innovative and changes the paradigm; it often looks "wrong" to reviewers in the old paradigm.

[unusual_whales @unusual_whales]

Americans now believe that college is not worth the cost, per NBC:

[Rohan Paul @rohanpaul_ai]

And that's Google's ‘Ironwood,’ its most powerful 7-th gen TPU.

- delivers 4,614 FP8 TFLOPS of performance and is equipped with 192 GB of HBM3E memory, offering a bandwidth of up to 7.37 TB/s.

- Ironwood pods scale up to 9,216 AI accelerators, delivering a total of 42.5 FP8 ExaFLOPS for training and inference, which by far exceeds the FP8 capabilities of Nvidia's GB300 NVL72 system that stands at 0.36 ExaFLOPS.

- The pod is interconnected using a proprietary 9.6 Tb/s Inter-Chip Interconnect network, and carries roughly 1.77 PB of HBM3E memory in total, once again exceeding what Nvidia's competing platform can offer.

---

From `Google Cloud` YT channel

[Turing Post (Ksenia Se) @TheTuringPost]

This Google paper presented at #NeurIPS2025 is a true gem.

In their search for a better backbone for sequence models, they:

• Reframe Transformers & RNNs as associative memory systems driven by attentional bias
• Reinterpret "forgetting" as retention regularization, not as erasure
• Combine these insights into Miras – a unified framework for designing next-gen sequence architectures

From this perspective, they introduce 3 new models, Moneta, Yaad, and Memora, that:

- Beat Transformers, Mamba2, DeltaNet, and hybrids across key benchmarks
- Scale better to long contexts
- Deliver state-of-the-art recall on needle-in-a-haystack tests

Here are the details (really worth exploring):

[Pedro Domingos @pmddomingos]

Unbelievable. China has a huge problem with scientific fraud, and when someone alludes to it the conversation is about how racist they are.

[Anastasis Stathopoulos @statho_]

Check out our new work "Score-Guided Diffusion for 3D Human Recovery", a.k.a. ScoreHMR, with @ligongh and Dimitris Metaxas that will appear at #CVPR2024!

Paper: https://t.co/lKMFBh2eow
Project Page: https://t.co/arDU2sMDKh
Code & models: https://t.co/xlO0FjYqKz

[Efthymios Tzinis @ETzinis]

This is the proper usage of machine learning models, removing communication barriers and enabling ideas of freedom to reach people's minds. Exceptional speech from Milei stating the obvious truths.

[Linus ✦ Ekenstam @LinusEkenstam]

Milei’s talk at Davos

This never happened.

But with AI (HeyGen) his voice is cloned and turned to English (Elevenlabs) and his mouth is re-synched to match the English words.

This is the true power of AI, anything can be consumed in any language

[Israel ישראל @Israel]

We debated whether or not to share these horrific images, but the world needs to know what we are up against.

These aren’t “freedom fighters”. They are Hamas terrorists, but no different than ISIS terrorists.

Same tactics, different names.

Butchering families, kidnapping grandmothers. Desecrating bodies.

Help us get this message out.

[Yuma Koizumi @yuma_koizumi]

Speech restoration method Miipher (used to generate LibriTTS-R) has been accepted to WASPAA!! It converts degraded speech to studio quality, and generates almost inexhaustible training data for speech generation.

Demo: https://t.co/nsrcIpIPaq
Paper: https://t.co/3D611h2E4C

[Jonathan Le Roux @JonathanLeRoux]

🥳🎉

[Jonathan Le Roux @JonathanLeRoux]

I will be presenting MERL's work on audio source separation and transcription during a seminar organized by #HiPARIScenter this Thursday, 6/22 @telecomparis and on Zoom.
@merl_news @IP_Paris_ @HECParis
Registration⬇️
https://t.co/VT5B3sPSzL

[Hao-Wen (Herman) Dong 董皓文 @hermanhwdong]

Excited to share that I will be presenting our audio synthesis paper at the CVPR Sight & Sound Workshop! 🤩

In this paper, we proposed a new text-to-audio synthesis model that can be trained using videos and a pretrained language-vision model.

#CVPR2023

https://t.co/1Q4PljvniO

[Efthymios Tzinis @ETzinis]

@arouditchenko @Google @GooglePixelFC Hey Andrew, I am in Cambridge as well! Just text me at the internal, see ya there!

[Efthymios Tzinis @ETzinis]

I am super excited to share that I just started as a Research Scientist @Google this week!🤩 Looking forward to delving deeper into audio-visual perception and building great products with the @GooglePixelFC folks!😃

[Jonathan Le Roux @JonathanLeRoux]

Darius's "Hyperbolic Audio Source Separation" paper has been recognized as one of the top 3% of all papers accepted at #ICASSP2023 @ieeeICASSP 👏🎉🍾
Paper: https://t.co/ggl93cdub1
Code: https://t.co/SujTNzkjT3
Demo video: https://t.co/rpqnWzuAL8

[Efthymios Tzinis @ETzinis]

@actuallyaswin Thanks so much Aswin!! Your self-supervised presentation was super as well!

[Efthymios Tzinis @ETzinis]

Feel free to stop by around 4pm to learn more about optimal condition training for sound separation! Higher upper bounds for conditional separation systems and state-of-the-art text-based separation performance! 🤩 @ieeeICASSP

[Efthymios Tzinis @ETzinis]

Our @ieeeICASSP23 optimal condition training presentation is online!https://t.co/v3UpyE5stp
Read our paper with @JonathanLeRoux,Gordon,Paris to learn how to train robust multi-condition separation systems + how to get SOTA text-based separation results!🧐https://t.co/fV95xTF2Gl