<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="waizman-etal-2026-erase">
    <titleInfo>
        <title>ERASE: Precise Machine Unlearning via Geometry-Aware Gradient Shaping and Adaptive Loss Control</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Itzik</namePart>
        <namePart type="family">Waizman</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Shahar</namePart>
        <namePart type="family">Katz</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Lior</namePart>
        <namePart type="family">Wolf</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>Large language models frequently memorize specific factual associations from their training data, necessitating post-hoc removal of targeted knowledge. However, standard gradient-based unlearning techniques treat all parameter directions uniformly, often leading to catastrophic forgetting of general capabilities. We introduce **ERASE** (**E**igenspace-**R**egulated **A**daptive **S**elective **E**rasure), a geometry-aware unlearning framework that combines three complementary mechanisms: (1) a *spectral filter* that attenuates gradient updates along principal directions of retained model knowledge, confining forgetting pressure to low-interference subspaces; (2) *selective error-signal construction*, which restricts gradients to semantically meaningful token positions, avoiding diffuse collateral damage; and (3) an *adaptive Cross-Entropy Target (CET) loss with dynamic per-sample stopping*, which drives each forget example toward a target cross-entropy and halts on convergence, preventing both under- and over-forgetting. Our method requires no reference model, no retain-set loss, and no auxiliary task supervision during training. On TOFU within the OpenUnlearning evaluation framework, ERASE achieves the highest aggregate score across four model families and two forget splits, attaining the best privacy and near-original utility among all compared baselines; on the document-level MUSE-News benchmark, with forget spans recovered automatically via named entity recognition, it simultaneously attains the strongest forgetting and retention with the privacy score closest to the retrain oracle. Our code is publicly available at https://github.com/ItzikWaizman/ERASE.</abstract>
    <identifier type="citekey">waizman-etal-2026-erase</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.2104/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>44796</start>
            <end>44824</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T ERASE: Precise Machine Unlearning via Geometry-Aware Gradient Shaping and Adaptive Loss Control
%A Waizman, Itzik
%A Katz, Shahar
%A Wolf, Lior
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F waizman-etal-2026-erase
%X Large language models frequently memorize specific factual associations from their training data, necessitating post-hoc removal of targeted knowledge. However, standard gradient-based unlearning techniques treat all parameter directions uniformly, often leading to catastrophic forgetting of general capabilities. We introduce **ERASE** (**E**igenspace-**R**egulated **A**daptive **S**elective **E**rasure), a geometry-aware unlearning framework that combines three complementary mechanisms: (1) a *spectral filter* that attenuates gradient updates along principal directions of retained model knowledge, confining forgetting pressure to low-interference subspaces; (2) *selective error-signal construction*, which restricts gradients to semantically meaningful token positions, avoiding diffuse collateral damage; and (3) an *adaptive Cross-Entropy Target (CET) loss with dynamic per-sample stopping*, which drives each forget example toward a target cross-entropy and halts on convergence, preventing both under- and over-forgetting. Our method requires no reference model, no retain-set loss, and no auxiliary task supervision during training. On TOFU within the OpenUnlearning evaluation framework, ERASE achieves the highest aggregate score across four model families and two forget splits, attaining the best privacy and near-original utility among all compared baselines; on the document-level MUSE-News benchmark, with forget spans recovered automatically via named entity recognition, it simultaneously attains the strongest forgetting and retention with the privacy score closest to the retrain oracle. Our code is publicly available at https://github.com/ItzikWaizman/ERASE.
%U https://aclanthology.org/2026.findings-emnlp.2104/
%P 44796-44824

Markdown (Informal)

[ERASE: Precise Machine Unlearning via Geometry-Aware Gradient Shaping and Adaptive Loss Control](https://aclanthology.org/2026.findings-emnlp.2104/) (Waizman et al., Findings 2026)

• ERASE: Precise Machine Unlearning via Geometry-Aware Gradient Shaping and Adaptive Loss Control (Waizman et al., Findings 2026)

ACL

• Itzik Waizman, Shahar Katz, and Lior Wolf. 2026. ERASE: Precise Machine Unlearning via Geometry-Aware Gradient Shaping and Adaptive Loss Control. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 44796–44824, Budapest, Hungary. Association for Computational Linguistics.