Advancing AI: Highlights from February

February at Sony AI was defined by momentum across global stages, research publications, and conversations about how AI moves from theory into practice.

This month spanned responsible AI conversations in India, upcoming keynote appearances in the U.S., new generative audio research headed to ICASSP 2026, and a deep discussion on how Sony approaches AI across gaming, music, film, and semiconductors. Together, these moments reflect a broader theme: connecting rigorous research with practical impact.

Here’s a look at the work that defined February.

Featured Sessions

Responsible AI in Practice: A Data Perspective | India AI Impact Summit 2026
Responsible AI cannot be retrofitted. It must be built by design.
At the India AI Impact Summit 2026, we brought together voices from research, industry, and law to talk about what that means in practice. The session opened with our work from the Sony AI Ethics team on FHIBE (Fair Human Image Benchmark for Evaluation), a consent-driven fairness benchmark published in Nature, and used that as a starting point to discuss data governance, evaluation, and accountability in real-world systems.
Shruti Nagpal (from Ethics team at Sony AI) moderated a panel with:

• -Dr. Mayank Vatsa, Professor at IIT Jodhpur
• -Ankur Saraswat, Vice President, AI Garage at Mastercard
• -Aprajita Rana, Partner at AZB & Partners

The discussion centred on critical questions facing the AI ecosystem today:

• -What does meaningful accountability look like inside large organizations?
• -Who should define and enforce standards for responsible data?
• -And how do legal and regulatory frameworks shape AI deployment and innovation, including the governance practices that are often overlooked?

“What stood out most during this event was a shared understanding that Responsible AI cannot be owned by any single function. It requires research, engineering, and governance to work together from day one, with data, evaluation, and accountability at the core,” notes Nagpal.

Sony AI researcher, Shruti Nagpal, presents FHIBE at the AI Impact Summit in New Delhi India.

Coming Soon

On March 3–4, at the Texas Symposium on Machine Learning, Responsible AI & Robotics at UT Austin, Alice Xiang will deliver a keynote examining the evolution of responsible AI as autonomous agents emerge—and why human-centric datasets remain foundational to building systems that are accountable and trusted.

For more information, visit: Texas Symposium on Machine Learning, Responsible AI & Robotics

And another reminder, Sony AI’s Chief Scientist, Peter Stone, will be in attendance at SXSW 2026, participating in a panel on humanoid robots and presenting on reinforcement learning.

Details can be found below:

Humanoid Robots Are Clocking In: Meet Who’s Putting Them To Work

Saturday March 14, 2026 | 11:30 a.m. CT | JW Marriott - Salon D

Is Reinforcement Learning the Real Future of AI?

Saturday, March 14, 2026 | 4 p.m. CT | JW Marriott - Salon AB

Sony AI President Michael Spranger joined Tiffany Kayo on Coral Capital (Episode #32) to discuss how Sony approaches artificial intelligence across its uniquely integrated ecosystem, from semiconductors to gaming, music, and film.

In the conversation, Spranger outlines how Sony AI builds differentiated AI alongside creators and product teams, and why responsible development remains foundational to the company’s long-term strategy. The episode offers an inside look at how research moves from the lab to real-world creative environments, and how trust, technical rigor, and cross-disciplinary collaboration shape Sony AI’s work.

Watch or listen to the full conversation:
YouTube
Spotify

Music Business Worldwide (MBW) published an in-depth analysis of Sony Group’s emerging approach to AI music detection and protection, highlighting recent research from Sony AI

The article examines three strands of work previously featured on the Sony AI blog:

Attribution — Research presented at NeurIPS 2025 introduces a method for identifying which training songs most influenced a generated output by selectively “unlearning” tracks from a model and observing measurable shifts. Rather than relying on surface-level similarity, the approach traces influence at the model level.

Read the Research: “Large-Scale Training Data Attribution for Music Generative Models via Unlearning,” accepted at the NeurIPS 2025 Creative AI Track, by Yuki Mitsufuji, Woosung Choi, Junghyun Koo, Kin Wai Cheuk, WeiHsiang Liao, and Joan Serrà.

Recognition — A paper accepted to ICML 2025 proposes segment-level matching through CLEWS, enabling detection of shared musical material in short audio excerpts. This fine-grained approach supports identifying near-duplicates and version relationships that traditional full-track matching systems may miss.

Read the Research: “Supervised Contrastive Learning from Weakly-Labeled Audio Segments for Musical Version Matching,” presented at ICML 2025 by Joan Serrà, R. Oguz Araz, Dmitry Bogdanov, and Yuki Mitsufuji.

Protection — Research accepted to INTERSPEECH 2025 evaluates how current watermarking systems perform against real-world transformations. Findings from RAW-Bench show that existing watermarking approaches struggle to survive AI-powered compression, raising important questions about how future authentication systems should be designed.

Read the Research: “A Comprehensive Real-World Assessment of Audio Watermarking Algorithms: Will They Survive Neural Codecs?”, presented at INTERSPEECH 2025, by Yigitcan Özer, Woosung Choi, Joan Serrà, Mayank Kumar Singh, Wei-Hsiang Liao, and Yuki Mitsufuji

The MBW piece frames these efforts as a layered technical blueprint: tracing influence in generative systems, mapping relationships between works, and stress-testing current protection methods. As noted in our earlier blog posts, this research is exploratory and research-driven, focused on strengthening musical integrity in the age of machine learning.

Read the full Music Business Worldwide article here: Sony Group’s blueprint for AI music detection tech is promising. Here’s what it’s working on…

Read the inspiration: Protecting Creator’s Rights in the Age of AI – Sony AI

ICASSP 2026 — Early Preview

The IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) is one of the longest-running venues for advances in audio, speech, and signal processing. The conference brings together research spanning foundational methods and practical systems that shape how sound is created and experienced.

We’re pleased to share that we have several papers accepted to ICASSP 2026, spanning ultra-low bitrate audio compression and generative approaches to music production.

We’ll return to both papers with deeper dives, implementation details, and author commentary as ICASSP approaches. Until then, please dive into our work from last year’s ICASSP: Unlocking the Future of Video-to-Audio Synthesis: Inside the MMAudio Model – Sony AI

Connect with us on LinkedIn, Instagram, or X, and let us know what you’d like to see in future editions. Until next month, keep imagining the possibilities with Sony AI.