<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="menon-ekbal-2026-far">
    <titleInfo>
        <title>How Far Do SAE Features Persist Across LLM Layers? An Empirical Study with an FFN Pruning Application</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Vaishakh</namePart>
        <namePart type="given">Sreekanth</namePart>
        <namePart type="family">Menon</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Asif</namePart>
        <namePart type="family">Ekbal</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>Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Dana</namePart>
            <namePart type="family">Arad</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Yonatan</namePart>
            <namePart type="family">Belinkov</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Hanjie</namePart>
            <namePart type="family">Chen</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Nils</namePart>
            <namePart type="family">Feldhus</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Aaron</namePart>
            <namePart type="family">Mueller</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Hosein</namePart>
            <namePart type="family">Mohebbi</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Gabriele</namePart>
            <namePart type="family">Sarti</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Martin</namePart>
            <namePart type="family">Tutek</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">979-8-89176-476-7</identifier>
    </relatedItem>
    <abstract>Sparse autoencoders (SAEs) are typically applied at the single layer they were trained on. We ask whether a pretrained SAE’s per-prompt feature activations persist when the same SAE is applied to residuals at deeper layers of the underlying language model, and whether the resulting cross-layer signal is strong enough to drive per-prompt FFN neuron pruning without any training or calibration data. Testing three SAE families on three roughly 8B-scale language models, we find adjacent-layer carry-through is high across families (0.72–0.82 at L_\textsrc+1) and its rate of decay differs qualitatively across sparsification mechanisms, with JumpReLU still activating half the features 17 layers downstream while TopK degrades within about five. We propose FARN (Feature and Anchor-based Removal of Neurons), a training-free per-prompt FFN pruner that scores each downstream neuron on the SAE’s decoder columns weighted by the current prompt’s active features, plus a prompt-independent anchor from the decoder bias. On six multiple-choice benchmarks, keeping roughly 17–20% of FFN neurons per layer, FARN closes the gap to an activation oracle (a baseline that peeks at the true downstream activations to pick the top-firing neurons) to within 1.5 percentage points and outperforms random, magnitude, and calibrated static-activation baselines on every model.</abstract>
    <identifier type="citekey">menon-ekbal-2026-far</identifier>
    <location>
        <url>https://aclanthology.org/2026.blackboxnlp-1.13/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>189</start>
            <end>200</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T How Far Do SAE Features Persist Across LLM Layers? An Empirical Study with an FFN Pruning Application
%A Menon, Vaishakh Sreekanth
%A Ekbal, Asif
%Y Arad, Dana
%Y Belinkov, Yonatan
%Y Chen, Hanjie
%Y Feldhus, Nils
%Y Mueller, Aaron
%Y Mohebbi, Hosein
%Y Sarti, Gabriele
%Y Tutek, Martin
%S Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-476-7
%F menon-ekbal-2026-far
%X Sparse autoencoders (SAEs) are typically applied at the single layer they were trained on. We ask whether a pretrained SAE’s per-prompt feature activations persist when the same SAE is applied to residuals at deeper layers of the underlying language model, and whether the resulting cross-layer signal is strong enough to drive per-prompt FFN neuron pruning without any training or calibration data. Testing three SAE families on three roughly 8B-scale language models, we find adjacent-layer carry-through is high across families (0.72–0.82 at L_\textsrc+1) and its rate of decay differs qualitatively across sparsification mechanisms, with JumpReLU still activating half the features 17 layers downstream while TopK degrades within about five. We propose FARN (Feature and Anchor-based Removal of Neurons), a training-free per-prompt FFN pruner that scores each downstream neuron on the SAE’s decoder columns weighted by the current prompt’s active features, plus a prompt-independent anchor from the decoder bias. On six multiple-choice benchmarks, keeping roughly 17–20% of FFN neurons per layer, FARN closes the gap to an activation oracle (a baseline that peeks at the true downstream activations to pick the top-firing neurons) to within 1.5 percentage points and outperforms random, magnitude, and calibrated static-activation baselines on every model.
%U https://aclanthology.org/2026.blackboxnlp-1.13/
%P 189-200

Markdown (Informal)

[How Far Do SAE Features Persist Across LLM Layers? An Empirical Study with an FFN Pruning Application](https://aclanthology.org/2026.blackboxnlp-1.13/) (Menon & Ekbal, BlackboxNLP 2026)

• How Far Do SAE Features Persist Across LLM Layers? An Empirical Study with an FFN Pruning Application (Menon & Ekbal, BlackboxNLP 2026)

ACL

• Vaishakh Sreekanth Menon and Asif Ekbal. 2026. How Far Do SAE Features Persist Across LLM Layers? An Empirical Study with an FFN Pruning Application. In Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 189–200, Budapest, Hungary. Association for Computational Linguistics.