arxiv: 2604.07316 · v1 · submitted 2026-04-08 · 💻 cs.LG

Recognition: 2 theorem links

· Lean Theorem

SL-FAC: A Communication-Efficient Split Learning Framework with Frequency-Aware Compression

Zehang Lin , Miao Yang , Haihan Zhu , Zheng Lin , Jianhao Huang , Jing Yang , Guangjin Pan , Dianxin Luan

show 4 more authors

Zihan Fang Shunzhi Zhu Wei Ni John Thompson

Authors on Pith no claims yet

Pith reviewed 2026-05-10 18:15 UTC · model grok-4.3

classification 💻 cs.LG

keywords split learningcommunication efficiencyfrequency decompositionquantization compressionsmashed dataedge computingdistributed training

0 comments

The pith

Transforming smashed data to the frequency domain and quantizing components by spectral energy lets split learning transmit far fewer bits while preserving the information needed for convergence.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Split learning partitions a neural network so that edge devices only handle the first layers, but this still requires sending large activation tensors called smashed data to the server and receiving gradients back. SL-FAC converts those tensors into the frequency domain, breaks them into spectral components that carry different amounts of information, and then applies coarser quantization to low-energy components while keeping finer precision on high-energy ones. The selective compression therefore shrinks the total bits exchanged without discarding the parts that most influence how the model updates its weights. Experiments show the resulting communication savings compared with earlier split-learning compression methods while reaching similar final accuracy and convergence speed. If the pattern holds, more devices could participate in collaborative training under tight bandwidth constraints.

Core claim

SL-FAC establishes that adaptive frequency decomposition of smashed data combined with frequency-based quantization compression delivers substantial communication savings in split learning by preserving high-energy spectral components that drive model convergence.

What carries the argument

Adaptive frequency decomposition (AFD) that transforms smashed data into the frequency domain and decomposes it into spectral components, paired with frequency-based quantization compression (FQC) that assigns bit widths according to each component's spectral energy distribution.

If this is right

Communication volume between edge devices and the server decreases while model convergence speed stays comparable to uncompressed split learning.
Training efficiency improves on resource-constrained edges because fewer bits are sent per round.
The same accuracy target is reached with lower total data exchanged across the tested neural network architectures.
More devices can join a single split-learning session without exhausting available bandwidth.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The frequency-aware principle could be tested in federated learning to see whether it reduces uplink costs there as well.
Energy distribution patterns might be stable enough across training runs to allow pre-computed quantization schedules that remove the need for per-round analysis.
Applying the decomposition only to certain layers or data modalities could yield further savings if high-energy components concentrate in predictable places.

Load-bearing premise

That selectively quantizing lower-energy frequency components will not introduce errors that meaningfully slow convergence or reduce final model accuracy across varied tasks and architectures.

What would settle it

A head-to-head run on a standard image-classification benchmark where SL-FAC reaches the same test accuracy and convergence speed as uncompressed split learning but with at least 50 percent less total communication volume; if accuracy drops or rounds needed increase, the central claim fails.

Figures

Figures reproduced from arXiv: 2604.07316 by Dianxin Luan, Guangjin Pan, Haihan Zhu, Jianhao Huang, Jing Yang, John Thompson, Miao Yang, Shunzhi Zhu, Wei Ni, Zehang Lin, Zheng Lin, Zihan Fang.

**Figure 2.** Figure 2: The training performance of SL-FAC on the MNIST and HAM10000 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

read the original abstract

The growing complexity of neural networks hinders the deployment of distributed machine learning on resource-constrained devices. Split learning (SL) offers a promising solution by partitioning the large model and offloading the primary training workload from edge devices to an edge server. However, the increasing number of participating devices and model complexity leads to significant communication overhead from the transmission of smashed data (e.g., activations and gradients), which constitutes a critical bottleneck for SL. To tackle this challenge, we propose SL-FAC, a communication-efficient SL framework comprising two key components: adaptive frequency decomposition (AFD) and frequency-based quantization compression (FQC). AFD first transforms the smashed data into the frequency domain and decomposes it into spectral components with distinct information. FQC then applies customized quantization bit widths to each component based on its spectral energy distribution. This collaborative approach enables SL-FAC to achieve significant communication reduction while strategically preserving the information most crucial for model convergence. Extensive experiments confirm the superior performance of SL-FAC for improving the training efficiency.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

SL-FAC applies frequency decomposition and energy-based quantization to smashed data in split learning, which is a straightforward engineering step that may cut communication but needs tighter experiment controls to confirm the accuracy tradeoff.

read the letter

SL-FAC introduces adaptive frequency decomposition followed by per-component quantization on the smashed data passed between client and server in split learning. The core move is to move the activations or gradients into the frequency domain, split them by spectral energy, and then assign fewer bits to low-energy parts. This is presented as a way to shrink the communication volume while keeping the parts that matter for convergence intact. The approach is new in how it frames the compression specifically around frequency content of the intermediate tensors rather than treating them as generic vectors. That pairing has not been the standard way prior SL compression papers handled the problem, so the framework itself is the incremental contribution. The paper does a clean job laying out the two modules and explaining why energy distribution serves as the importance signal. If the full results hold up, the method could be practical for edge setups where bandwidth is the limiter. The main soft spot is that the abstract and high-level description leave the experimental controls thin. We do not see the exact baselines (uniform quantization, other frequency methods, or simple pruning), the datasets, the model sizes, or the statistical significance of the accuracy numbers. Without those, it is difficult to judge whether the claimed lack of convergence slowdown is robust or tied to particular choices. The assumption that spectral energy is a reliable proxy for task-critical information is plausible but not automatic across architectures or loss landscapes. A reader working on distributed training for constrained devices would find the framework description useful and might pull the compression idea for their own pipeline. The work is coherent on its own terms and shows clear engagement with the split-learning bottleneck, so it merits sending to referees who can verify the implementation and run additional checks on the tradeoff curves.

Referee Report

0 major / 3 minor

Summary. The paper proposes SL-FAC, a split learning framework with two components: adaptive frequency decomposition (AFD), which transforms smashed data (activations/gradients) into the frequency domain and decomposes it into spectral components, and frequency-based quantization compression (FQC), which assigns per-component quantization bit widths according to spectral energy distribution. The central claim is that this yields substantial communication savings while preserving information critical for convergence, with extensive experiments asserted to demonstrate superior training efficiency over baselines.

Significance. If the empirical results hold across varied tasks and architectures, the approach could meaningfully alleviate the communication bottleneck in split learning for resource-constrained edge devices, offering a frequency-domain alternative to standard compression techniques. The energy-based bit allocation provides a principled heuristic that may generalize beyond the evaluated settings.

minor comments (3)

Abstract: the claim of 'extensive experiments confirm the superior performance' is stated without any quantitative results, dataset names, baseline methods, or accuracy/communication metrics, which weakens the reader's ability to gauge the strength of the evidence for the central claim.
The description of FQC does not specify the exact procedure for computing spectral energy per component or the mapping from energy to bit-width allocation; an equation or algorithm box would clarify whether this is fully deterministic or involves tunable thresholds.
The weakest assumption—that frequency decomposition and selective quantization introduce no meaningful convergence slowdown or accuracy loss—would benefit from explicit discussion of failure cases or sensitivity analysis in the experiments section.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their careful review and positive recommendation of minor revision. The referee's summary accurately captures the core elements of SL-FAC, including the roles of adaptive frequency decomposition (AFD) and frequency-based quantization compression (FQC) in reducing communication overhead while preserving convergence-critical information. No major comments were raised in the report.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces SL-FAC as an empirical framework combining adaptive frequency decomposition (AFD) and frequency-based quantization compression (FQC) to reduce communication in split learning. The approach transforms smashed data to the frequency domain and allocates bits by spectral energy, with performance claims resting on experimental validation rather than any closed-form derivation or mathematical prediction. No equations reduce inputs to outputs by construction, no parameters are fitted and then relabeled as predictions, and no load-bearing self-citations or uniqueness theorems are invoked. The central claims remain independently testable via accuracy and communication metrics on external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The proposal rests on standard signal-processing assumptions applied to ML activations; no new entities invented and limited free parameters visible from abstract.

free parameters (1)

Per-component quantization bit widths
Chosen based on spectral energy distribution; likely tuned per experiment though not quantified in abstract.

axioms (1)

domain assumption Frequency-domain components of smashed data carry separable information content that can be selectively compressed without harming overall model convergence
Invoked by the design of AFD followed by FQC.

pith-pipeline@v0.9.0 · 5510 in / 1087 out tokens · 105532 ms · 2026-05-10T18:15:07.744456+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

AFD transforms the smashed data into the frequency domain and decouples it into components... FQC then applies customized quantization bit widths to each component based on its spectral energy distribution.
IndisputableMonolith/Foundation/AlphaCoordinateFixation.lean J_uniquely_calibrated_via_higher_derivative unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

the cumulative energy ratio R^t_{c,(k)} ... energy threshold θ ... low-frequency F_l ... high-frequency F_h

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages

[1]

Efficient multi- user resource allocation for urban vehicular edge computing: a hybrid architecture matching approach,

H. Xie, H. Liu, H. Chen, S. Feng, Z. Wei, and Y . Zeng, “Efficient multi- user resource allocation for urban vehicular edge computing: a hybrid architecture matching approach,”IEEE Trans. V eh. Technol., vol. 74, no. 1, pp. 1811–1816, 2024

work page 2024
[2]

Hi- erarchical Split Federated Learning: Convergence Analysis and System Optimization,

Z. Lin, W. Wei, Z. Chen, C.-T. Lam, X. Chen, Y . Gao, and J. Luo, “Hi- erarchical Split Federated Learning: Convergence Analysis and System Optimization,”IEEE Trans. Mobile Comput., 2025

work page 2025
[3]

Automated Federated Pipeline for Parameter-Efficient Fine-Tuning of Large Lan- guage Models,

Z. Fang, Z. Lin, Z. Chen, X. Chen, Y . Gao, and Y . Fang, “Automated Federated Pipeline for Parameter-Efficient Fine-Tuning of Large Lan- guage Models,”IEEE Trans. Mobile Comput., 2025

work page 2025
[4]

State-aware perturbation optimization for robust deep reinforcement learning,

Z. Zhang, T. Duan, Z. Lin, D. Huang, Z. Fang, Z. Sun, L. Xiong, H. Liang, H. Cui, and Y . Cui, “State-aware perturbation optimization for robust deep reinforcement learning,”IEEE Trans. Mobile Comput., 2025

work page 2025
[5]

Leed: A highly efficient and scalable llm-empowered expert demonstrations framework for multi-agent rein- forcement learning,

T. Duan, Z. Zhang, S. Guo, D. Huang, Y . Zhao, Z. Lin, Z. Fang, D. Luan, H. Cui, and Y . Cui, “Leed: A highly efficient and scalable llm-empowered expert demonstrations framework for multi-agent rein- forcement learning,”arXiv preprint arXiv:2509.14680, 2025

work page arXiv 2025
[6]

Efficient Parallel Split Learning over Resource-Constrained Wireless Edge Networks,

Z. Lin, G. Zhu, Y . Deng, X. Chen, Y . Gao, K. Huang, and Y . Fang, “Efficient Parallel Split Learning over Resource-Constrained Wireless Edge Networks,”IEEE Trans. Mobile Comput., vol. 23, no. 10, pp. 9224–9239, 2024

work page 2024
[7]

Hfedmoe: Resource-aware heterogeneous federated learning with mixture-of-experts,

Z. Fang, Z. Lin, S. Hu, Y . Ma, Y . Tao, Y . Deng, X. Chen, and Y . Fang, “Hfedmoe: Resource-aware heterogeneous federated learning with mixture-of-experts,”arXiv preprint arXiv:2601.00583, 2026

work page arXiv 2026
[8]

Rrto: A high-performance transparent offloading system for model inference in mobile edge computing,

Z. Sun, X. Guan, Z. Lin, Y . Qing, H. Song, Z. Fang, Z. Chen, F. Liu, H. Cui, W. Niet al., “Rrto: A high-performance transparent offloading system for model inference in mobile edge computing,”arXiv preprint arXiv:2507.21739, 2025

work page arXiv 2025
[9]

Satfed: A resource-efficient leo satellite-assisted heterogeneous federated learning framework,

Y . Zhang, Z. Lin, Z. Chen, Z. Fang, W. Zhu, X. Chen, J. Zhao, and Y . Gao, “Satfed: A resource-efficient leo satellite-assisted heterogeneous federated learning framework,”Engineering, 2024

work page 2024
[10]

LLM-Driven Stationarity-Aware Expert Demonstrations for Multi-Agent Reinforcement Learning in Mobile Systems,

T. Duan, Z. Zhang, Z. Lin, S. Guo, X. Guan, G. Wu, Z. Fang, H. Meng, X. Du, J.-Z. Zhouet al., “LLM-Driven Stationarity-Aware Expert Demonstrations for Multi-Agent Reinforcement Learning in Mobile Systems,”arXiv preprint arXiv:2511.19368, 2025

work page arXiv 2025
[11]

IC3M: In-Car Multimodal Multi-Object Monitoring for Abnormal Status of Both Driver and Passengers,

Z. Fang, Z. Lin, S. Hu, H. Cao, Y . Deng, X. Chen, and Y . Fang, “IC3M: In-Car Multimodal Multi-Object Monitoring for Abnormal Status of Both Driver and Passengers,”arXiv preprint arXiv:2410.02592, 2024

work page arXiv 2024
[12]

Distributed Resource Allocation with Federated Learning for Delay- Sensitive IoV Services,

X. Song, Y . Hua, Y . Yang, G. Xing, F. Liu, L. Xu, and T. Song, “Distributed Resource Allocation with Federated Learning for Delay- Sensitive IoV Services,”IEEE Trans. V eh. Technol., vol. 73, no. 3, pp. 4326–4336, 2023

work page 2023
[13]

Adaptsfl: Adaptive Split Federated Learning in Resource-Constrained Edge Networks,

Z. Lin, G. Qu, W. Wei, X. Chen, and K. K. Leung, “Adaptsfl: Adaptive Split Federated Learning in Resource-Constrained Edge Networks,” IEEE Trans. Netw., 2025

work page 2025
[14]

Accelerating Federated Learning with Model Segmentation for Edge Networks,

M. Hu, J. Zhang, X. Wang, S. Liu, and Z. Lin, “Accelerating Federated Learning with Model Segmentation for Edge Networks,”IEEE Trans. Green Commun. Netw., 2024

work page 2024
[15]

Conflict-aware client selection for multi-server federated learning,

M. Hong, Z. Lin, Z. Lin, L. Li, M. Yang, X. Du, Z. Fang, Z. Kang, D. Luan, and S. Zhu, “Conflict-aware client selection for multi-server federated learning,”arXiv preprint arXiv:2602.02458, 2026

work page arXiv 2026
[16]

Aggregation alignment for federated learning with mixture-of-experts under data heterogeneity,

Z. Fang, Q. Wang, H. An, Z. Lin, Y . Deng, X. Chen, and Y . Fang, “Aggregation alignment for federated learning with mixture-of-experts under data heterogeneity,”arXiv preprint arXiv:2603.21276, 2026

work page arXiv 2026
[17]

Fedsn: A federated learning framework over heterogeneous leo satellite networks,

Z. Lin, Z. Chen, Z. Fang, X. Chen, X. Wang, and Y . Gao, “Fedsn: A federated learning framework over heterogeneous leo satellite networks,” IEEE Trans. Mobile Comput., vol. 24, no. 3, pp. 1293–1307, 2024

work page 2024
[18]

Demystifying Small Language Models for Edge Deployment,

Z. Lu, X. Li, D. Cai, R. Yi, F. Liu, W. Liu, J. Luan, X. Zhang, N. D. Lane, and M. Xu, “Demystifying Small Language Models for Edge Deployment,” inProc. ACL, 2025, pp. 14 747–14 764

work page 2025
[19]

Split Learning in 6G Edge Networks,

Z. Lin, G. Qu, X. Chen, and K. Huang, “Split Learning in 6G Edge Networks,”IEEE Wirel. Commun., 2024

work page 2024
[20]

Optimal resource allocation for u-shaped parallel split learning,

S. Lyu, Z. Lin, G. Qu, X. Chen, X. Huang, and P. Li, “Optimal resource allocation for u-shaped parallel split learning,” in2023 IEEE Globecom Workshops (GC Wkshps), 2023, pp. 197–202. 6

work page 2023
[21]

Nsc-sl: A bandwidth-aware neural subspace compression for communication-efficient split learning,

Z. Fang, M. Yang, Z. Lin, Z. Lin, Z. Fang, Z. Zhang, T. Duan, D. Huang, and S. Zhu, “Nsc-sl: A bandwidth-aware neural subspace compression for communication-efficient split learning,”arXiv preprint arXiv:2602.02696, 2026

work page arXiv 2026
[22]

HASFL: Heterogeneity- aware Split Federated Learning over Edge Computing Systems,

Z. Lin, Z. Chen, X. Chen, W. Ni, and Y . Gao, “HASFL: Heterogeneity- aware Split Federated Learning over Edge Computing Systems,”IEEE Trans. Mobile Comput., 2026

work page 2026
[23]

Exploiting Label-Aware Channel Scoring for Adaptive Channel Pruning in Split Learning,

J. Tan, Z. Lin, X. Cai, R. Zhu, Z. Fang, P. Chen, and W. Ni, “Exploiting Label-Aware Channel Scoring for Adaptive Channel Pruning in Split Learning,”arXiv preprint arXiv:2603.09792, 2026

work page arXiv 2026
[24]

HSplitLoRA: A Heterogeneous Split Parameter- Efficient Fine-Tuning Framework for Large Language Models,

Z. Lin, Y . Zhang, Z. Chen, Z. Fang, X. Chen, P. Vepakomma, W. Ni, J. Luo, and Y . Gao, “HSplitLoRA: A Heterogeneous Split Parameter- Efficient Fine-Tuning Framework for Large Language Models,”arXiv preprint arXiv:2505.02795, 2025

work page arXiv 2025
[25]

Reducing Communication for Split Learning by Randomized Top-k Sparsification,

F. Zheng, C. Chen, L. Lyu, and B. Yao, “Reducing Communication for Split Learning by Randomized Top-k Sparsification,” inProc. IJCAI, 2023, pp. 4665–4673

work page 2023
[26]

BottleNet: A Deep Learn- ing Architecture for Intelligent Mobile Cloud Computing Services,

A. E. Eshratifar, A. Esmaili, and M. Pedram, “BottleNet: A Deep Learn- ing Architecture for Intelligent Mobile Cloud Computing Services,” in Proc. ISLPED, Jul. 2019, pp. 1–6

work page 2019
[27]

Communication-Efficient Split Learning via Adaptive Feature-Wise Compression,

Y . Oh, J. Lee, C. G. Brinton, and Y .-S. Jeon, “Communication-Efficient Split Learning via Adaptive Feature-Wise Compression,”IEEE Trans. Neural Netw. Learn. Syst., vol. 36, no. 6, pp. 10 844–10 858, Jan. 2025

work page 2025
[28]

2 P- Encoder: On Exploration of Channel-Class Correlation for Multi-Label Zero-Shot Learning,

Z. Liu, S. Guo, X. Lu, J. Guo, J. Zhang, Y . Zeng, and F. Huo, “2 P- Encoder: On Exploration of Channel-Class Correlation for Multi-Label Zero-Shot Learning,” inProc. CVPR, 2023, pp. 23 859–23 868

work page 2023
[29]

Channel Interaction Networks for Fine-Grained Image Categorization,

Y . Gao, X. Han, X. Wang, W. Huang, and M. Scott, “Channel Interaction Networks for Fine-Grained Image Categorization,” inProc. AAAI, 2020, pp. 10 818–10 825

work page 2020
[30]

Multimodal-oriented Interactive Joint Source-channel Coding for Lightweight Semantic Com- munication,

X. Niu, L. Tan, J. Wu, W. Yuan, and T. Q. Quek, “Multimodal-oriented Interactive Joint Source-channel Coding for Lightweight Semantic Com- munication,”IEEE Trans. V eh. Technol., vol. 74, no. 10, pp. 16 516– 16 520, Oct. 2025

work page 2025
[31]

3D point clouds data super resolution-aided LiDAR odometry for vehicular positioning in urban canyons,

J. Yue, W. Wen, J. Han, and L.-T. Hsu, “3D point clouds data super resolution-aided LiDAR odometry for vehicular positioning in urban canyons,”IEEE Trans. V eh. Technol., vol. 70, no. 5, pp. 4098–4112, 2021

work page 2021
[32]

DCTMamba: Advancing JPEG Image Restoration Through Long-Sequence Modeling and Adaptive Frequency Strategy,

X. Wang, X. Fu, L. Li, and Z.-J. Zha, “DCTMamba: Advancing JPEG Image Restoration Through Long-Sequence Modeling and Adaptive Frequency Strategy,” inProc. AAAI, 2025, pp. 7925–7933

work page 2025
[33]

Harmonic Convolutional Networks Based on Discrete Cosine Transform,

M. Ulicny, V . A. Krylov, and R. Dahyot, “Harmonic Convolutional Networks Based on Discrete Cosine Transform,”Pattern Recogn., vol. 129, p. 108707, Sep. 2022

work page 2022
[34]

Multi-objective Convex Quantization for Efficient Model Compression,

C. Fan, D. Guo, Z. Wang, and M. Wang, “Multi-objective Convex Quantization for Efficient Model Compression,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 47, no. 4, pp. 2313–2329, Apr. 2025

work page 2025
[35]

STanH: Parametric Quantization for Variable Rate Learned Image Compression,

A. Presta, E. Tartaglione, A. Fiandrotti, and M. Grangetto, “STanH: Parametric Quantization for Variable Rate Learned Image Compression,” IEEE Trans. Image Process., vol. 34, no. 4, pp. 639–651, Jan. 2025

work page 2025
[36]

NLIC: Non-uniform Quantization-based Learned Image Compression,

Z. Ge, S. Ma, W. Gao, J. Pan, and C. Jia, “NLIC: Non-uniform Quantization-based Learned Image Compression,”IEEE Trans. Circuits Syst. Video Technol., vol. 34, no. 10, pp. 9647–9663, Oct. 2024

work page 2024
[37]

Differential Privacy Decen- tralized Federated Learning for Internet of Vehicles over Time-varying Unbalanced Networks,

Z. Zhao, Z. Liu, C. Zhang, and M. Wei, “Differential Privacy Decen- tralized Federated Learning for Internet of Vehicles over Time-varying Unbalanced Networks,”IEEE Trans. V eh. Technol., 2025

work page 2025
[38]

The HAM10000 Dataset, A Large Collection of Multi-source Dermatoscopic Images of Common Pigmented Skin Lesions,

P. Tschandl, C. Rosendahl, and H. Kittler, “The HAM10000 Dataset, A Large Collection of Multi-source Dermatoscopic Images of Common Pigmented Skin Lesions,”Sci. Data, vol. 5, no. 1, pp. 1–9, Aug. 2018

work page 2018
[39]

PowerQuant: Auto- morphism Search for Non-Uniform Quantization,

E. Yvinec, A. Dapogny, M. Cord, and K. Bailly, “PowerQuant: Auto- morphism Search for Non-Uniform Quantization,” inProc. ICLR, 2023, pp. 1–21

work page 2023
[40]

EasyQuant: An Efficient Data-free Quantization Algorithm for LLMs,

H. Tang, Y . Sun, D. Wu, K. Liu, J. Zhu, and Z. Kang, “EasyQuant: An Efficient Data-free Quantization Algorithm for LLMs,” inProc. EMNLP, 2023, pp. 9119–9128

work page 2023