@inproceedings{dershowitz-bouneffouf-2026-controlled,
title = "Controlled Fine-Tuning for Two-Tower Recommenders: Why Gradual Unfreezing Can Fail and Cosine Regularization Suffices",
author = "Dershowitz, Gabriel and
Bouneffouf, Djallel",
editor = "Sedoc, Jo{\~a}o and
Shu, Raphael and
Drozd, Aleksandr and
Akula, Arjun and
Rogers, Anna and
Giorgi, Salvatore and
Benton, Adrian",
booktitle = "Proceedings of the Seventh Workshop on Insights from Negative Results in {NLP}",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.insights-1.5/",
pages = "43--47",
ISBN = "979-8-89176-480-4",
abstract = "Controlled fine-tuning of pretrained text encoders {---} via gradual layer-wise unfreezing and anchoring {---} is widely assumed to benefit downstream tasks. We test this assumption in two-tower recommendation and findthree negative results. First, any fine-tuning dramatically beats a frozen encoder ({\ensuremath{\geq}} 2.7{\texttimes}cold Recall@10 lift); freezing the item tower is strongly counterproductive. Second, gradual unfreezing worsens performance compared to vanilla fine-tuning under realistic budgets (12epochs). Third, a fixed cosine drift penalty is the only consistently beneficial component; our full adaptive recipe does not outperform vanilla fine-tuning. Thus, for two-tower retrieval, practitioners should default to vanilla fine-tuning with light regularization {---} gradual unfreezing is unnecessary and harmful, challenging conventional wisdom from language model fine-tuning."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dershowitz-bouneffouf-2026-controlled">
<titleInfo>
<title>Controlled Fine-Tuning for Two-Tower Recommenders: Why Gradual Unfreezing Can Fail and Cosine Regularization Suffices</title>
</titleInfo>
<name type="personal">
<namePart type="given">Gabriel</namePart>
<namePart type="family">Dershowitz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Djallel</namePart>
<namePart type="family">Bouneffouf</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 Seventh Workshop on Insights from Negative Results in NLP</title>
</titleInfo>
<name type="personal">
<namePart type="given">João</namePart>
<namePart type="family">Sedoc</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Raphael</namePart>
<namePart type="family">Shu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aleksandr</namePart>
<namePart type="family">Drozd</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Arjun</namePart>
<namePart type="family">Akula</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anna</namePart>
<namePart type="family">Rogers</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Salvatore</namePart>
<namePart type="family">Giorgi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Adrian</namePart>
<namePart type="family">Benton</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">979-8-89176-480-4</identifier>
</relatedItem>
<abstract>Controlled fine-tuning of pretrained text encoders — via gradual layer-wise unfreezing and anchoring — is widely assumed to benefit downstream tasks. We test this assumption in two-tower recommendation and findthree negative results. First, any fine-tuning dramatically beats a frozen encoder (\ensuremath\geq 2.7×cold Recall@10 lift); freezing the item tower is strongly counterproductive. Second, gradual unfreezing worsens performance compared to vanilla fine-tuning under realistic budgets (12epochs). Third, a fixed cosine drift penalty is the only consistently beneficial component; our full adaptive recipe does not outperform vanilla fine-tuning. Thus, for two-tower retrieval, practitioners should default to vanilla fine-tuning with light regularization — gradual unfreezing is unnecessary and harmful, challenging conventional wisdom from language model fine-tuning.</abstract>
<identifier type="citekey">dershowitz-bouneffouf-2026-controlled</identifier>
<location>
<url>https://aclanthology.org/2026.insights-1.5/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>43</start>
<end>47</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Controlled Fine-Tuning for Two-Tower Recommenders: Why Gradual Unfreezing Can Fail and Cosine Regularization Suffices
%A Dershowitz, Gabriel
%A Bouneffouf, Djallel
%Y Sedoc, João
%Y Shu, Raphael
%Y Drozd, Aleksandr
%Y Akula, Arjun
%Y Rogers, Anna
%Y Giorgi, Salvatore
%Y Benton, Adrian
%S Proceedings of the Seventh Workshop on Insights from Negative Results in NLP
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-480-4
%F dershowitz-bouneffouf-2026-controlled
%X Controlled fine-tuning of pretrained text encoders — via gradual layer-wise unfreezing and anchoring — is widely assumed to benefit downstream tasks. We test this assumption in two-tower recommendation and findthree negative results. First, any fine-tuning dramatically beats a frozen encoder (\ensuremath\geq 2.7×cold Recall@10 lift); freezing the item tower is strongly counterproductive. Second, gradual unfreezing worsens performance compared to vanilla fine-tuning under realistic budgets (12epochs). Third, a fixed cosine drift penalty is the only consistently beneficial component; our full adaptive recipe does not outperform vanilla fine-tuning. Thus, for two-tower retrieval, practitioners should default to vanilla fine-tuning with light regularization — gradual unfreezing is unnecessary and harmful, challenging conventional wisdom from language model fine-tuning.
%U https://aclanthology.org/2026.insights-1.5/
%P 43-47
Markdown (Informal)
[Controlled Fine-Tuning for Two-Tower Recommenders: Why Gradual Unfreezing Can Fail and Cosine Regularization Suffices](https://aclanthology.org/2026.insights-1.5/) (Dershowitz & Bouneffouf, insights 2026)
ACL