Investigating the interpretability of biometric face templates using gated sparse autoencoders and differentiable image parametrizations

Rot, Peter (avtor), Grm, Klemen (avtor)

State-of-the-art face recognition models rely on deep, complex neural net architectures that produce relatively compact template vectors, making their mechanisms of operation difficult to interpret and understand. Recently, mechanistic interpretability has emerged as a promising approach to explain large language models. In this paper, we aim to apply such approaches to explain face recognition models. Our method involves transforming face image templates into sparse representations and analyzing their components by identifying images that maximize activation. Our results demonstrate that existing mechanistic interpretability techniques generalize well to previously unconsidered tasks and architectures, and that differentiable image parametrizations can serve as a useful additional means of confirming the interpretation of sparse representations.

• Jezik: Angleški jezik
• Ključne besede: biometry, face recognition, sparse auto-encoders
• Tipologija: 1.08 - Objavljeni znanstveni prispevek na konferenci
• Organizacija: FE - Fakulteta za elektrotehniko
• Status publikacije: Objavljeno
• Različica publikacije: Objavljena publikacija
• Leto izida: 2024
• Št. strani: 5 str.
• PID: 20.500.12556/RUL-160088
• UDK: 004.93'1
• COBISS.SI-ID: 204387843
• Datum objave v RUL: 19.08.2024

Gradivo je del monografije

• Naslov: ICML 2024 Workshop on Mechanistic Interpretability : ICML 2024 MI Workshop
• Kraj izida: [Massachusetts
• Založnik: OpenReview
• Leto izida: 2024
• COBISS.SI-ID: 204384771

Licence

• Licenca: CC BY 4.0, Creative Commons Priznanje avtorstva 4.0 Mednarodna
• Povezava: http://creativecommons.org/licenses/by/4.0/deed.sl
• Opis: To je standardna licenca Creative Commons, ki daje uporabnikom največ možnosti za nadaljnjo uporabo dela, pri čemer morajo navesti avtorja.

Sekundarni jezik

• Jezik: Slovenski jezik
• Ključne besede: biometrija, razpoznavanje obrazov, redki samo-kodirniki

Projekti

• Financer: ARIS - Javna agencija za znanstvenoraziskovalno in inovacijsko dejavnost Republike Slovenije
• Številka projekta: P2-0250
• Naslov: Metrologija in biometrični sistemi

• Financer: ARIS - Javna agencija za znanstvenoraziskovalno in inovacijsko dejavnost Republike Slovenije
• Številka projekta: J2-50069
• Naslov: Interpretacija mehanizmov za razložljivo biometrično umetno inteligenco (MIXBAI)