Yuan-Yen Tai, Ph.D. · Machine-learning scientist

From theoretical physics to machine learning.

I started in theoretical physics and mathematics, and have spent the past decade researching and engineering scalable solutions for high-complexity systems. From Amazon to Apple, I have taken five patented machine-learning innovations from concept to deployment, in large-scale infrastructure and on edge devices alike.

Yuan-Yen Tai standing between two large letter Ys, one blue on yellow and one yellow on blue
5patented ML innovations
19scientific publications
10+years of research

Path

  1. Feb 2026 – Present

    The Biological Computing Co.

    • Built the adapter model behind the company's first public, real-time interactive demo. Added to the open-source Oasis video model, it uses under 3% more parameters, keeps long outputs stable, and holds quality at fewer diffusion steps while running several times faster. Read the blog post ↗ · Try the demo ↗
    • Trained a new video generation model on the company's in-house bio-derived adapter: diffusion runs several times faster with no loss in output quality. Trained on a multi-GPU cluster.
    • Now building the third high-performance adapter and leading a team across training, benchmark validation, quantization and efficiency optimization.
  2. 2021 – 2026

    Apple

    • Led the development and launch of an end-to-end agentic RAG product: a hybrid chain of four models, each above 95% accuracy.
    • Invented and filed a patent for an on-screen codex with a computer-vision model.
    • Designed, implemented and patented a hybrid Graph + LLM search and ranking algorithm.
  3. 2019 – 2021

    CargoMetrics

    • Led development of large-scale data pre-processing algorithms on the PySpark framework.
    • Led and designed the model for classifying and tracking cargo-ship movements.
  4. 2017 – 2019

    Amazon

    • Invented an audio watermark algorithm that broke through the physical limits of transmitting a watermark signal through the air.
    • Lead scientist for the on-device and backend audio watermark encoding / decoding algorithm.
  5. 2014 – 2017

    Los Alamos National Lab Postdoctoral researcher

    • Multidimensional quantum-physics modeling research with GPU-accelerated computation.

Side projects

Audionyx

Audio identity for the GenAI era: a high-performance audio sign encoding / decoding system, from the device to the backend.

audionyx.ai ↗

AOWL

Audio Watermark Lab: invisible signatures that make a voice traceable to its origin.

aowl.app ↗

Avengents

One shared room for people and AI coding agents across machines, on iPhone, iPad, Mac and the web. Many colleagues at my company now use it to join their agents and finish projects together, and work faster for it.

avengents.ai ↗

Also: neuron digitization, an algorithm for pre- and post-trained LLM modeling.

Patents

Publications

Machine learning & signal processing 2017 – 2023

  1. Yuan-Yen Tai et al., Automatic App Image Generation and Selection. Review in progress for SIGIR 2024; related patent submitted (2023).
  2. Yuan-Yen Tai et al., Large-Language-Model boosted Graph structure for personalized-PageRank recommendation. Review in progress; related patent submitted (2023).
  3. Yuan-Yen Tai et al., Recreating and Specializing PageRank for Online Services. Apple internal ML summit (2022).
  4. Yuan-Yen Tai et al., Fast and scalable geometric clustering algorithm for privacy enhanced ML applications. Apple internal ML summit (2022).
  5. Yuan-Yen Tai and Mohamed F. Mansour, Audio watermarking over the air with modulated self-correlation. ICASSP 2019 (2019).
  6. Yuan-Yen Tai, An efficient clustering algorithm from the measure of local Gaussian distribution. arXiv:1709.08470 (2017).

Condensed-matter physics 2006 – 2019
  1. Hongchul Choi, Yuan-Yen Tai and Jian-Xin Zhu, Strong-Correlation Derived Spin-Fermion Model for Skyrmions in MnGe. PRB 99, 134437 (2019).
  2. Yuan-Yen Tai and Jian-Xin Zhu, Local charge variation enhanced spin canting of BiFeO3/La0.7Sr0.3MnO3 heterostructure. arXiv:1603.03107.
  3. Yuan-Yen Tai, H. Choi, T. Ahmed, C. S. Ting and J.-X. Zhu, Edge states and local electronic structure around an adsorbed impurity in a topological superconductor. PRB 92, 174514 (2015).
  4. Yuan-Yen Tai, C.-C. J. Wang, M. J. Graf, J.-X. Zhu and C. S. Ting, Emergent topological mirror insulator in t2g-orbital systems. PRB 91, 041111(R) (2015).
  5. Y. Zhao, Yuan-Yen Tai and C. S. Ting, Phase diagram of the isovalent phosphorous-substituted 122-type iron pnictides. PRB 91, 205110 (2015).
  6. Bo Li, L. Pan, Yuan-Yen Tai, M. J. Graf, J.-X. Zhu, K. E. Bassler and C. S. Ting, A unified description of superconducting pairing symmetry in electron-doped Fe-based-122 compounds. PRB 90, 220509(R) (2015).
  7. Y. Fang, Yuan-Yen Tai, J. Deng, C. Wu, X. Ding, J. Sun and E. K. H. Salje, Supercond. Sci. Technol. 28, 095004 (2015).
  8. L. Pan, J. Li, Yuan-Yen Tai, M. J. Graf, J.-X. Zhu and C. S. Ting, Evolution of quasiparticle states with and without a Zn impurity in doped 122 iron pnictides. PRB 90, 134501 (2014).
  9. L. Pan, J. Li, Yuan-Yen Tai, M. J. Graf, J.-X. Zhu and C. S. Ting, Evolution of the Fermi surface topology in doped 122 iron pnictides. PRB 88, 214510 (2013).
  10. H. Chen, Yuan-Yen Tai, C. S. Ting, M. J. Graf, J. Dai and J.-X. Zhu, Disorder effects in multiorbital s±-wave superconductors: Implications for Zn-doped BaFe2As2 compounds. PRB 88, 184509 (2013).
  11. Yuan-Yen Tai, J.-X. Zhu, M. J. Graf and C. S. Ting, Calculated phase diagram of doped BaFe2As2 superconductor in a C4-symmetry breaking model. EPL 103, 67001 (2013).
  12. Yuan-Yen Tai, J.-X. Zhu, M. J. Graf and C. S. Ting, Fermi surface evolution and checker-board block-spin antiferromagnetism in AxFe2-ySe2. PRB 86, 134512 (2012).
  13. Yuan-Yen Tai, P.-C. Li and H.-C. Tseng, A two-dimensional modified Levy-walk model for the DNA sequences. Physica A 369, 688 (2006).

Talks & education

  • 2019Lecture speaker, IEEE ICASSP 2019, Brighton, UK
  • 2018Invited speaker, audio summit, Amazon Lab126, Sunnyvale
  • 2018Live demo of watermark over the air, Q2 All-Hands, Amazon Boston
  • 2016Invited speaker, Brookhaven National Laboratory
  • 2016Oral presentation, APS March Meeting, Baltimore
  • 2015BLABS seminar talk, LANL T-4 Group
  • 2015Oral presentation, CNLS 35th Annual Conference
  • 2015Oral presentation, APS March Meeting, San Antonio
  • 2014Oral presentation, APS March Meeting, Denver
  • 2013Oral presentation, APS March Meeting, Baltimore
  • 2013Oral presentation, TcSUH Semiannual Symposium
  • 2012Oral presentation, APS March Meeting, Boston

Education

Ph.D., Physics

Condensed Matter Theory
University of Houston, Houston, TX · 2014

1st Place · TcSUH Research Symposium