---
OA_place: publisher
OA_type: gold
_id: '20821'
abstract:
- lang: eng
  text: Modern deep neural networks exhibit heterogeneity across numerous layers of
    various types such as residuals, multi-head attention, etc., due to varying structures
    (dimensions, activation functions, etc.), distinct representation characteristics,
    which impact predictions. We develop a general layer-wise quantization framework
    with tight variance and code-length bounds, adapting to the heterogeneities over
    the course of training. We then apply a new layer-wise quantization technique
    within distributed variational inequalities (VIs), proposing a novel Quantized
    Optimistic Dual Averaging (QODA) algorithm with adaptive learning rates, which
    achieves competitive convergence rates for monotone VIs. We empirically show that
    QODA achieves up to a 150% speedup over the baselines in end-to-end training time
    for training Wasserstein GAN on 12+GPUs.
acknowledgement: "This work was supported by Hasler Foundation Program: Hasler Responsible
  AI (project number 21043). The research was also sponsored by the Army Research
  Office and was accomplished under Grant Number W911NF-24-1-0048. This work was further
  funded by the Swiss National Science Foundation (SNSF) under grant number 200021_205011.
  We also acknowledge project A11 of the Swiss National Supercomputing Centre (CSCS)
  for providing computing resources. Dan Alistarh and Ilia Markov were supported in
  part through the ERC Proofof-Concept grant FastML (Grant Agreement 101158077). Ali
  Ramezani-Kebrya was supported by the Research Council of Norway through FRIPRO Grant
  under project number 356103, its Centres of Excellence scheme, Integreat - Norwegian
  Centre for knowledge-driven machine learning under\r\nproject number 332645 - and
  its Centre for Research-based Innovation funding scheme (Visual Intelligence under
  grant no. 309439)."
alternative_title:
- PMLR
article_processing_charge: No
arxiv: 1
author:
- first_name: Anh Duc
  full_name: Nguyen, Anh Duc
  last_name: Nguyen
- first_name: Ilia
  full_name: Markov, Ilia
  id: D0CF4148-C985-11E9-8066-0BDEE5697425
  last_name: Markov
- first_name: Frank Zhengqing
  full_name: Wu, Frank Zhengqing
  last_name: Wu
- first_name: Ali
  full_name: Ramezani-Kebrya, Ali
  last_name: Ramezani-Kebrya
- first_name: Kimon
  full_name: Antonakopoulos, Kimon
  last_name: Antonakopoulos
- first_name: Dan-Adrian
  full_name: Alistarh, Dan-Adrian
  id: 4A899BFC-F248-11E8-B48F-1D18A9856A87
  last_name: Alistarh
  orcid: 0000-0003-3650-940X
- first_name: Volkan
  full_name: Cevher, Volkan
  last_name: Cevher
citation:
  ama: 'Nguyen AD, Markov I, Wu FZ, et al. Layer-wise quantization for quantized optimistic
    dual averaging. In: <i>42nd International Conference on Machine Learning</i>.
    Vol 267. ML Research Press; 2025:46026-46072.'
  apa: 'Nguyen, A. D., Markov, I., Wu, F. Z., Ramezani-Kebrya, A., Antonakopoulos,
    K., Alistarh, D.-A., &#38; Cevher, V. (2025). Layer-wise quantization for quantized
    optimistic dual averaging. In <i>42nd International Conference on Machine Learning</i>
    (Vol. 267, pp. 46026–46072). Vancouver, Canada: ML Research Press.'
  chicago: Nguyen, Anh Duc, Ilia Markov, Frank Zhengqing Wu, Ali Ramezani-Kebrya,
    Kimon Antonakopoulos, Dan-Adrian Alistarh, and Volkan Cevher. “Layer-Wise Quantization
    for Quantized Optimistic Dual Averaging.” In <i>42nd International Conference
    on Machine Learning</i>, 267:46026–72. ML Research Press, 2025.
  ieee: A. D. Nguyen <i>et al.</i>, “Layer-wise quantization for quantized optimistic
    dual averaging,” in <i>42nd International Conference on Machine Learning</i>,
    Vancouver, Canada, 2025, vol. 267, pp. 46026–46072.
  ista: 'Nguyen AD, Markov I, Wu FZ, Ramezani-Kebrya A, Antonakopoulos K, Alistarh
    D-A, Cevher V. 2025. Layer-wise quantization for quantized optimistic dual averaging.
    42nd International Conference on Machine Learning. ICML: International Conference
    on Machine Learning, PMLR, vol. 267, 46026–46072.'
  mla: Nguyen, Anh Duc, et al. “Layer-Wise Quantization for Quantized Optimistic Dual
    Averaging.” <i>42nd International Conference on Machine Learning</i>, vol. 267,
    ML Research Press, 2025, pp. 46026–72.
  short: A.D. Nguyen, I. Markov, F.Z. Wu, A. Ramezani-Kebrya, K. Antonakopoulos, D.-A.
    Alistarh, V. Cevher, in:, 42nd International Conference on Machine Learning, ML
    Research Press, 2025, pp. 46026–46072.
conference:
  end_date: 2025-07-19
  location: Vancouver, Canada
  name: 'ICML: International Conference on Machine Learning'
  start_date: 2025-07-13
date_created: 2025-12-14T23:02:06Z
date_published: 2025-05-01T00:00:00Z
date_updated: 2025-12-16T12:46:54Z
day: '01'
ddc:
- '000'
department:
- _id: DaAl
external_id:
  arxiv:
  - '2505.14371'
file:
- access_level: open_access
  checksum: a7edf0e4304171a3e035842b3aab1704
  content_type: application/pdf
  creator: dernst
  date_created: 2025-12-16T12:45:41Z
  date_updated: 2025-12-16T12:45:41Z
  file_id: '20830'
  file_name: 2025_ICML_Nguyen.pdf
  file_size: 756213
  relation: main_file
  success: 1
file_date_updated: 2025-12-16T12:45:41Z
has_accepted_license: '1'
intvolume: '       267'
language:
- iso: eng
license: https://creativecommons.org/licenses/by/4.0/
month: '05'
oa: 1
oa_version: Published Version
page: 46026-46072
project:
- _id: 8e35c14b-16d5-11f0-9cad-a3fc35339161
  grant_number: '101158077'
  name: 'FastML: Efficient and Cost-Effective Distributed Machine Learning'
publication: 42nd International Conference on Machine Learning
publication_identifier:
  eissn:
  - 2640-3498
publication_status: published
publisher: ML Research Press
quality_controlled: '1'
scopus_import: '1'
status: public
title: Layer-wise quantization for quantized optimistic dual averaging
tmp:
  image: /images/cc_by.png
  legal_code_url: https://creativecommons.org/licenses/by/4.0/legalcode
  name: Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)
  short: CC BY (4.0)
type: conference
user_id: 2DF688A6-F248-11E8-B48F-1D18A9856A87
volume: 267
year: '2025'
...
---
OA_place: repository
OA_type: green
_id: '21858'
abstract:
- lang: eng
  text: "The recent surge in high-quality open-source Generative AI text models (colloquially:
    LLMs), as well as efficient finetuning techniques, have opened the possibility
    of creating high-quality personalized models that generate text attuned to a specific
    individual’s needs and are capable of credibly imitating their writing style by
    refining an open-source model with that person’s own data. The technology to create
    such models is accessible to private individuals, and training and running such
    models can be done cheaply on consumer-grade hardware. While these advancements
    are a huge gain for usability and privacy, this position paper argues that the
    practical feasibility of impersonating specific individuals also introduces novel
    safety risks. For instance, this technology enables the creation of phishing emails\r\nor
    fraudulent social media accounts, based on small amounts of publicly available
    text, or by the individuals themselves to escape AI text detection. We further
    argue that these risks are complementary to—and distinct from—the much-discussed
    risks of other impersonation attacks such as image, voice, or video deepfakes,
    and are not adequately addressed by the larger research community, or the current
    generation of open- and closed-source models."
acknowledgement: "This research was supported by the Scientific Service Units (SSU)
  of IST Austria through resources\r\nprovided by Scientific Computing (SciComp).
  EI was supported in part by the FWF DK VGSCO,\r\ngrant agreement number W1260-N35.
  AJ was supported in part by ERC Proof-of-Concept Grant\r\nFastML, grant agreement
  101158077."
article_processing_charge: No
arxiv: 1
author:
- first_name: Eugenia B
  full_name: Iofinova, Eugenia B
  id: f9a17499-f6e0-11ea-865d-fdf9a3f77117
  last_name: Iofinova
  orcid: 0000-0002-7778-3221
- first_name: Andrej
  full_name: Jovanovic, Andrej
  last_name: Jovanovic
- first_name: Dan-Adrian
  full_name: Alistarh, Dan-Adrian
  id: 4A899BFC-F248-11E8-B48F-1D18A9856A87
  last_name: Alistarh
  orcid: 0000-0003-3650-940X
citation:
  ama: 'Iofinova EB, Jovanovic A, Alistarh D-A. Position: It’s time to act on the
    risk of efficient personalized text generation. <i>arXiv</i>. doi:<a href="https://doi.org/10.48550/arXiv.2502.06560">10.48550/arXiv.2502.06560</a>'
  apa: 'Iofinova, E. B., Jovanovic, A., &#38; Alistarh, D.-A. (n.d.). Position: It’s
    time to act on the risk of efficient personalized text generation. <i>arXiv</i>.
    <a href="https://doi.org/10.48550/arXiv.2502.06560">https://doi.org/10.48550/arXiv.2502.06560</a>'
  chicago: 'Iofinova, Eugenia B, Andrej Jovanovic, and Dan-Adrian Alistarh. “Position:
    It’s Time to Act on the Risk of Efficient Personalized Text Generation.” <i>ArXiv</i>,
    n.d. <a href="https://doi.org/10.48550/arXiv.2502.06560">https://doi.org/10.48550/arXiv.2502.06560</a>.'
  ieee: 'E. B. Iofinova, A. Jovanovic, and D.-A. Alistarh, “Position: It’s time to
    act on the risk of efficient personalized text generation,” <i>arXiv</i>. .'
  ista: 'Iofinova EB, Jovanovic A, Alistarh D-A. Position: It’s time to act on the
    risk of efficient personalized text generation. arXiv, <a href="https://doi.org/10.48550/arXiv.2502.06560">10.48550/arXiv.2502.06560</a>.'
  mla: 'Iofinova, Eugenia B., et al. “Position: It’s Time to Act on the Risk of Efficient
    Personalized Text Generation.” <i>ArXiv</i>, doi:<a href="https://doi.org/10.48550/arXiv.2502.06560">10.48550/arXiv.2502.06560</a>.'
  short: E.B. Iofinova, A. Jovanovic, D.-A. Alistarh, ArXiv (n.d.).
corr_author: '1'
date_created: 2026-05-11T08:55:23Z
date_published: 2025-06-02T00:00:00Z
date_updated: 2026-05-19T11:20:27Z
day: '02'
department:
- _id: GradSch
- _id: DaAl
doi: 10.48550/arXiv.2502.06560
external_id:
  arxiv:
  - '2502.06560'
language:
- iso: eng
main_file_link:
- open_access: '1'
  url: https://doi.org/10.48550/arXiv.2502.06560
month: '06'
oa: 1
oa_version: Preprint
project:
- _id: 8e35c14b-16d5-11f0-9cad-a3fc35339161
  grant_number: '101158077'
  name: 'FastML: Efficient and Cost-Effective Distributed Machine Learning'
- _id: 9B9290DE-BA93-11EA-9121-9846C619BF3A
  grant_number: W1260-N35
  name: Vienna Graduate School on Computational Optimization
publication: arXiv
publication_status: draft
related_material:
  record:
  - id: '21854'
    relation: dissertation_contains
    status: public
status: public
title: 'Position: It''s time to act on the risk of efficient personalized text generation'
type: preprint
user_id: 8b945eb4-e2f2-11eb-945a-df72226e66a9
year: '2025'
...