Stefan Grafberger
@sgrafberger
Ph.D. Student at @bifoldberlin, researching data management for ML
ID: 1044475802824450048
https://stefan-grafberger.com 25-09-2018 06:36:51
126 Tweet
344 Followers
416 Following
Today I started my research internship in Redmond with the Microsoft Microsoft Gray Systems Lab. Looking forward to an amazing summer!
How fast can we unlearn data from recommendation models for the #righttobeforgotten? We unlearn interactions from kNN models on 1m users & 250m interactions with sub-second latency in our upcoming short paper SIGIR 2024 dl.acm.org/doi/10.1145/35… Joint work w/ Mozhdeh Ariannezhad Maarten de Rijke
Awesome paper by xiaozhong lyu Stefan Grafberger @ce__zhang Sebastian shows how #RAG can be improved through data importance learning. The approach learns weights for data sources based on their performance on a validation set and then re-weights or prunes the corpus. 1/3
Going to VLDB 2025 🇬🇧 ? Don't miss our demo on "mlwhatif: What If You Could Stop Re-Implementing Your Machine Learning Pipeline Analyses Over and Over?" Joint work with Stefan Grafberger and Shubha vldb.org/pvldb/vol16/p4…
Checkout our vision on "Red Onions, Soft Cheese and Data: From Food Safety to Data Traceability for Responsible AI" sites.computer.org/debull/A24mar/… We ask what reponsible data management can learn from the existing regulations and processes for food safety (w/ Stefan Grafberger & Ce Zhang)
Joint work with Stefan Grafberger Zhang Zeyu and Ce Zhang !
Our paper "Towards Interactively Improving ML Data Preparation Code via 'Shadow Pipelines'" has been accepted for the DEEM Workshop @ SIGMOD at SIGMOD! 🎉 In this vision paper, we present our initial ideas for my next research project. Joint work with Sebastian and Paul Groth.
LLMs show a huge potential for data wrangling. But how we can we increase their parameter-, compute- and sample efficiency for real-world use cases? Zhang Zeyu will present our ideas for this at the DBML workshop at #icde2024 IEEE ICDE Conference today! wis.ewi.tudelft.nl/assets/files/d…
We can't wait for DEEM Workshop @ SIGMOD 2024 to get started @sigmodconf! Join us tomorrow Sunday 9 June from 9am in the Tupungato room!! Check out the full program at: deem-workshop.github.io 🇨🇱
Looking forward to presenting our vision "Towards Interactively Improving ML Data Preparation Code via 'Shadow Pipelines'" at the DEEM Workshop @ SIGMOD today! The talk will be around 10:40 a.m. in the Tupungato room. stefan-grafberger.com/shadow-pipelin…
📢 Excited to share Feature Clock, an open-source library and paper accepted at IEEE VIS! Feature Clock enhances the explainability and compactness of visualizations of high-dimensional effects in two-dimensional plots. Big thanks to my co-authors Valentina Boeva and Rita Sevastjanova!