Talk by Chhavi Yadav (CMU)

on YouTube ↗17m

" summary

Study interviews 21 interviewees from banks, hospitals, platform teams and researchers and finds a clear misalignment: papers rarely address the top practitioner concerns of awareness and infrastructure, while real projects prioritize data quantity over quality, use public or synthetic data for prototyping, and lean on legal contracts as primary trust. Platforms like Flower and NVIDIA FLARE reduce deployment friction, but open problems remain: pre-training data valuation, opt-out guarantees, model ownership and post-deployment maintenance.

AI-written summary · about

" chapters (11)

1. 0:00Cross-silo FL challenges
2. 1:09FL lacks mainstream adoption
3. 3:44Ideation stage challenges
4. 5:02Prototyping and data quantity
5. 5:51Evaluation and legal contracts
6. 7:20Incentives and network limits
7. 8:33Data valuation issues
8. 8:54Deployment and model ownership
9. 10:24Research versus practice gap
10. 11:06Future directions
11. 12:35Summary and Q&A

" related talks

• 03mInterviewLukas Biewald · Scale By The Bay 2018
• 06mInterviewAdelbert Chang · Scale By The Bay 2019
• 31mShyam Sundaram: Head Digital Innovations & Transformation at Bajaj Finance at Cypher 2017Shyam Sundaram · Cypher 2017
• 14mIndustrial practitioners' mental models of adversarial machine learningKatharina Krombholz, Michael Backes, Lukas Bieringer, Kathrin Grosse, Battista Biggio · SOUPS 2022
• 16mOperationalizing AI in Finance: Real-Time AnalyticsMallikarjun Ramasani · Conf42 MLOps 2025
• 05mRaj Kamath interviewedRaj Kamath · Strata Data Conference 2017 - New York, New York
• 13m"We are not Future-ready": Understanding AI Privacy Risks and Existing Mitigation...Patrick Gage Kelley, Sai Teja Peddinti, Kurt Thomas, Alexandra Klymenko, Stephen Meisenbacher, Florian Matthes · SOUPS 2025
• 44mPanel Discussion: The BFSI Playbook - Scaling AI with Trust and QualityEshwar Pastapur, Brianna Elsass, Brian Corkery, Anand Moorthy · TestMu Conference 2026