Laura Kopf
@lkopf_ml
Followers
63
Following
37
Media
14
Statuses
32
PhD student in Interpretable Machine Learning @bifoldberlin @TUBerlin
Berlin
Joined May 2024
🔍 When do neurons encode multiple concepts?. We introduce PRISM, a framework for extracting multi-concept feature descriptions to better understand polysemanticity. 📄 Capturing Polysemanticity with PRISM: A Multi-Concept Feature Description Framework.🧵
1
4
13
Grateful to the institutions that supported this work:.@TUBerlin.@bifoldberlin.@UMI_Lab_AI.@FraunhoferHHI.@unipotsdam.@LeibnizATB. (7/7).
0
0
2
Many thanks to my amazing co-authors:.@nfelnlp.@kirill_bykov.@BommerPhiline.@anna_hedstroem.@Marina_MCV.@EberleOliver. (6/7).
1
0
2
Our results highlight that the PRISM framework not only provides multiple human interpretable descriptions for neurons but also aligns with the human interpretation of polysemanticity. (5/7)
1
0
2
In exploring the concept space, we use PRISM to characterize more complex components, finding and interpreting patterns that specific attention heads or groups of neurons respond to. (4/7)
1
0
3
We benchmark PRISM across layers and architectures, showing how polysemanticity and interpretability shift through the model. (3/7)
1
0
2
PRISM samples sentences from the top percentile activation distribution, clusters them in embedding space, and uses an LLM to generate labels for each concept cluster. (2/7)
1
0
3
Huge thanks to my incredible supervisor @kirill_bykov, who laid the foundation for this project and provided brilliant guidance 🙏, and to @BommerPhiline and @SLapuschkin, who unfortunately couldn’t be there.
0
0
2
Still overwhelmed by the amazing response to our poster session at @NeurIPSConf with @anna_hedstroem and @Marina_MCV! It was incredible to have such lively and inspiring discussions with brilliant people whose work I admire. ✨
2
3
6
Want to know more about CoSy?.đź“„ Paper: đź’» Code: đź”— Poster:
1
0
0
Special thanks to our supporting institutions:.@UMI_Lab_AI, @bifoldberlin, @ml_tuberlin, @TUBerlin, @unipotsdam, @LeibnizATB, and @FraunhoferHHI.
1
0
2
My co-authors @anna_hedstroem and @Marina_MCV will also be at @NeurIPSConf. A big thank you to my other co-authors @kirill_bykov, @BommerPhiline, and @SLapuschkin, who unfortunately couldn’t be there.
1
0
1
I’ll be presenting our work at @NeurIPSConf in Vancouver! 🎉.Join me this Thursday, December 12th, in East Exhibit Hall A-C, Poster #3107, from 11 a.m. PST to 2 p.m. PST. I'll be discussing our paper “CoSy: Evaluating Textual Explanations of Neurons.”
1
4
16
RT @kirill_bykov: I am not attending #NeurIPS2024, but I encourage everyone interested in #XAI and #MechInterp to check out our paper on ev….
0
3
0
RT @NeelNanda5: NeurIPS has an overwhelming amount of papers, so I made myself a hacky spreadsheet of all (well, most) of the interpretabil….
0
51
0
Special thanks to our supporting institutions: @UMI_Lab_AI.@unipotsdam.@LeibnizATB.@bifoldberlin.@FraunhoferHHI.@TUBerlin . #InterpretableML #MechInterp #ExplainableAI.
0
0
3
🎉 Excited to announce that our paper has been accepted to #NeurIPS2024! This is my first first-author publication 🥳 I'm incredibly grateful to my amazing supervisor @kirill_bykov and co-authors @BommerPhiline @anna_hedstroem @SLapuschkin @Marina_MCV!. 📄
1
3
20
Join us today at the #ICML2024 Workshop on the Next Generation of AI Safety! Find @kirill_bykov and me in Hall A1 at Poster Session #2, from 3:30 PM to 4:30 PM. Looking forward to seeing you there!
0
3
13
RT @UMI_Lab_AI: Join us at the @icmlconf in Vienna next week. We are presenting two of our papers at the Mechanistic Interpretability and N….
0
1
0