pith. machine review for the scientific record. sign in

arxiv: 2605.05819 · v1 · submitted 2026-05-07 · 💻 cs.LG

Recognition: unknown

HCInfer: An Efficient Inference System via Error Compensation for Resource-Constrained Devices

Shen Xu , Xiangwen Zhuge , Zhe Xu , Yingkun Hu , Zheng Yang , Yunhao Liu
Authors on Pith no claims yet

Pith reviewed 2026-05-08 14:39 UTC · model grok-4.3

classification 💻 cs.LG
keywords LLM inferencemodel compressionerror compensationLoRA branchesheterogeneous computingresource-constrained devicesasynchronous pipeline
0
0 comments X

The pith

HCInfer offloads error-compensation branches to the CPU while running compressed LLMs on the GPU to recover accuracy without major speed loss.

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

The paper establishes that error compensation branches in compressed LLMs can be offloaded to the CPU because they access only a small subset of parameters each step. It introduces an asynchronous pipeline and sensitivity-aware dynamic rank allocation to hide the added latency. This produces higher accuracy than pure compression while delivering substantial speedups over full-precision inference. A sympathetic reader would care because the approach targets the practical barrier of running large models on everyday consumer hardware.

Core claim

HCInfer runs a compressed backbone on the GPU and offloads residual LoRA-style compensation branches to the CPU, using an asynchronous compensation pipeline and sensitivity-aware dynamic rank allocation to hide overhead and maximize accuracy recovery on resource-constrained devices.

What carries the argument

The asynchronous compensation pipeline that executes compressed model layers on GPU while offloading error compensation to CPU, combined with dynamic rank allocation based on parameter sensitivity.

If this is right

  • Accuracy on downstream tasks improves by a maximum of 5.2 percent relative to the compressed model alone.
  • Inference speed reaches a maximum of 10.4 times that of the full-precision model.
  • Deployment of large models becomes practical on memory-limited consumer devices without severe accuracy loss or throughput collapse.

Where Pith is reading between the lines

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

  • The same offloading pattern could apply to other compensation or adaptation modules beyond LoRA branches.
  • Combining HCInfer with quantization might further reduce GPU memory while preserving the accuracy gains.
  • The dynamic rank allocation could be tested on tasks with different sensitivity profiles to check robustness.

Load-bearing premise

Compensation branches access only a small subset of parameters per inference step and the asynchronous CPU-GPU pipeline fully hides communication and computation latency without creating new bottlenecks.

What would settle it

A timing profile on target consumer hardware showing that CPU-GPU transfers for the compensation branches add more than a small fraction of total inference time, eliminating the claimed speedup.

Figures

Figures reproduced from arXiv: 2605.05819 by Shen Xu, Xiangwen Zhuge, Yingkun Hu, Yunhao Liu, Zheng Yang, Zhe Xu.

Figure 1
Figure 1. Figure 1: Overview of different inference paradigms. view at source ↗
Figure 2
Figure 2. Figure 2: Singular value distribution of quantization error matrices for various modules in layer 0 of Qwen-30B-A3B, based on GPTQ Q4 quantization. Existing studies have shown that quantization error matri￾ces often exhibit strong low-rank structure [11, 12]. How￾ever, our analysis reveals that the usefulness of compensa￾tion varies substantially across different components of the model. Singular-value distribution.… view at source ↗
Figure 3
Figure 3. Figure 3: Matrix-wise output sensitivity analysis of Qwen-30B-A3B under GPTQ Q4 quantization. view at source ↗
Figure 4
Figure 4. Figure 4: Overview of HCInfer. Based on the observations in § 3.1, the quantization error matrix ∆W exhibits strong low-rank structure, shown as: Y = XWˆ + X∆W ≈ XWˆ + (XAr)Br, where Ar ∈ R d×r and Br ∈ R r×rk are low-rank factors obtained from the singular value decompo￾sition of ∆W, and r ≪ min(d, k) denotes the compensation rank. 4.2 Heterogeneous Compensation Pipeline To make full use of computing resources on r… view at source ↗
Figure 5
Figure 5. Figure 5: End-to-end throughput and accuracy comparison of HCInfer across models and datasets. view at source ↗
Figure 6
Figure 6. Figure 6: End-to-end performance comparison for various input and output length combinations. view at source ↗
Figure 7
Figure 7. Figure 7: Throughput of HCInfer w/ and w/o het￾erogeneous pipeline. The baseline is a naive of￾floading scheme where all compensation weights are offloaded to CPU DRAM and moved to GPU only when accessed. Pipeline Ablation: To verify the contribution of heterogeneous parallel computing, we compared HCInfer with a naive compensation parameter offloading scheme view at source ↗
Figure 8
Figure 8. Figure 8: Throughput and accuracy comparison of HCInfer w/ and w/o dynamic rank allocator. view at source ↗
Figure 9
Figure 9. Figure 9: Detailed illustration of the heterogeneous execution pipeline. The purple blocks in (b) view at source ↗
read the original abstract

LLMs often struggle with memory-constrained deployment on consumer-grade hardware due to their massive parameter sizes. While existing solutions such as model compression and offloading improve deployment feasibility, they often suffer from substantial accuracy degradation or severe throughput bottlenecks. Recent error compensation methods recover accuracy through auxiliary LoRA-style branches, and we observe that these branches are inherently amenable to offloading: they require substantial parameter storage but access only a small subset of compensation parameters during each inference step. Motivated by this opportunity, we propose HCInfer, a heterogeneous inference system that offloads residual compensation to the CPU while executing the compressed backbone on the GPU, and further introduces an asynchronous compensation pipeline and sensitivity-aware dynamic rank allocation to hide compensation overhead and maximize accuracy recovery. Experimental results show that HCInfer achieves a maximum accuracy improvement of 5.2% on downstream tasks compared to compression model and sustaining a maximum speedup of 10.4x compared to full-precision model.

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.

Referee Report

1 major / 1 minor

Summary. The manuscript presents HCInfer, a heterogeneous inference system for deploying large language models on resource-constrained devices. It offloads LoRA-style error compensation branches to the CPU while running the compressed model backbone on the GPU. Key innovations include an asynchronous compensation pipeline and sensitivity-aware dynamic rank allocation to minimize overhead and maximize accuracy recovery. The authors report experimental results showing up to 5.2% accuracy improvement over compressed baselines and up to 10.4x speedup over full-precision models.

Significance. This work tackles the important challenge of efficient LLM inference on consumer hardware by combining model compression with targeted error compensation offloading. The observation that compensation branches access only a small subset of parameters per step is leveraged effectively for heterogeneous execution. If the performance claims hold under scrutiny, the approach could have practical significance for edge deployment of LLMs, offering a balance between accuracy and speed not easily achieved by compression or offloading alone.

major comments (1)
  1. [Abstract] The headline claims of a maximum 5.2% accuracy improvement and 10.4x speedup (Abstract) depend on the asynchronous CPU-GPU compensation pipeline fully hiding both computation and PCIe transfer latency. No per-layer timing breakdown, no ablation with the pipeline disabled, and no verification that dynamic rank allocation keeps the per-step parameter footprint small are described, leaving the core overlap assumption untested and the speedup result difficult to evaluate.
minor comments (1)
  1. [Abstract] The abstract would benefit from briefly naming the specific downstream tasks, model architectures, and compression baselines used to achieve the reported numbers.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the practical significance of HCInfer for edge LLM deployment. We address the major comment on experimental validation of the asynchronous pipeline below. We will incorporate the requested analyses in the revised manuscript to make the performance claims more robust.

read point-by-point responses

  1. Referee: [Abstract] The headline claims of a maximum 5.2% accuracy improvement and 10.4x speedup (Abstract) depend on the asynchronous CPU-GPU compensation pipeline fully hiding both computation and PCIe transfer latency. No per-layer timing breakdown, no ablation with the pipeline disabled, and no verification that dynamic rank allocation keeps the per-step parameter footprint small are described, leaving the core overlap assumption untested and the speedup result difficult to evaluate.

    Authors: We agree that the manuscript would benefit from additional validation of the latency-hiding mechanism. The reported 5.2% accuracy improvement is obtained by comparing the final compensated model against the compressed baseline on downstream tasks and does not depend on the execution pipeline. The 10.4x speedup, however, does rely on effective overlap between GPU backbone execution and CPU compensation (including PCIe transfers). In the revision we will add: (1) per-layer timing breakdowns that quantify GPU compute time, CPU compensation time, and the achieved overlap; (2) an ablation that disables the asynchronous pipeline (i.e., synchronous CPU-GPU execution) and reports the resulting throughput degradation; (3) explicit measurements of the per-step compensation parameter footprint under sensitivity-aware dynamic rank allocation, confirming that only a small subset of parameters is accessed per inference step. These additions will directly substantiate the overlap assumption and allow readers to evaluate the speedup claims more rigorously. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical system measurements with no self-referential derivations

full rationale

The paper describes a heterogeneous inference system (HCInfer) using offloading of LoRA-style compensation branches to CPU, an asynchronous pipeline, and sensitivity-aware rank allocation. All load-bearing claims are experimental: measured accuracy recovery (max 5.2%) and speedup (max 10.4x) versus baselines. No equations, fitted parameters, or derivations appear that reduce by construction to their own inputs. The design choices are motivated by observation but validated externally via timing and accuracy benchmarks on downstream tasks; no self-citation chains or ansatzes underpin the central results.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical model or parameters are presented; the paper is an engineering systems contribution.

pith-pipeline@v0.9.0 · 5472 in / 973 out tokens · 50848 ms · 2026-05-08T14:39:08.642576+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 15 canonical work pages · 5 internal anchors

  1. [1]

    From automation to autonomy: A survey on large language models in scientific discovery

    Tianshi Zheng, Zheye Deng, Hong Ting Tsang, Weiqi Wang, Jiaxin Bai, Zihao Wang, and Yangqiu Song. From automation to autonomy: A survey on large language models in scientific discovery. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 17744–17761, 2025

    2025

  2. [2]

    A survey on llm-based multi-agent systems: workflow, infrastructure, and challenges.Vicinagearth, 1(1):9, 2024

    Xinyi Li, Sai Wang, Siqi Zeng, Yu Wu, and Yi Yang. A survey on llm-based multi-agent systems: workflow, infrastructure, and challenges.Vicinagearth, 1(1):9, 2024

    2024

  3. [3]

    Foundation models and intelligent decision-making: Progress, challenges, and perspectives.The Innovation, 6(6), 2025

    Jincai Huang, Yongjun Xu, Qi Wang, Qi Cheems Wang, Xingxing Liang, Fei Wang, Zhao Zhang, Wei Wei, Boxuan Zhang, Libo Huang, et al. Foundation models and intelligent decision-making: Progress, challenges, and perspectives.The Innovation, 6(6), 2025

    2025

  4. [4]

    Qwen3 Technical Report

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    work page internal anchor Pith review arXiv 2025

  5. [5]

    GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers

    Elias Frantar, Saleh Ashkboos, Torsten Hoefler, and Dan Alistarh. Gptq: Accurate post-training quantization for generative pre-trained transformers.arXiv preprint arXiv:2210.17323, 2022

    work page internal anchor Pith review arXiv 2022

  6. [6]

    Awq: Activation-aware weight quantization for on-device llm compression and acceleration.Proceedings of machine learning and systems, 6:87–100, 2024

    Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei-Chen Wang, Guangxuan Xiao, Xingyu Dang, Chuang Gan, and Song Han. Awq: Activation-aware weight quantization for on-device llm compression and acceleration.Proceedings of machine learning and systems, 6:87–100, 2024

    2024

  7. [7]

    arXiv:2407.11062 , year=

    Mengzhao Chen, Wenqi Shao, Peng Xu, Jiahao Wang, Peng Gao, Kaipeng Zhang, and Ping Luo. Efficientqat: Efficient quantization-aware training for large language models.arXiv preprint arXiv:2407.11062, 2024

    work page arXiv 2024

  8. [8]

    Flexgen: High-throughput generative inference of large language models with a single gpu

    Ying Sheng, Lianmin Zheng, Binhang Yuan, Zhuohan Li, Max Ryabinin, Beidi Chen, Percy Liang, Christopher Ré, Ion Stoica, and Ce Zhang. Flexgen: High-throughput generative inference of large language models with a single gpu. InInternational Conference on Machine Learning, pages 31094–31116. PMLR, 2023

    2023

  9. [9]

    Hetegen: Efficient heterogeneous parallel inference for large language models on resource-constrained devices.Proceedings of Machine Learning and Systems, 6:162–172, 2024

    Xuanlei Zhao, Bin Jia, Haotian Zhou, Ziming Liu, Shenggan Cheng, and Yang You. Hetegen: Efficient heterogeneous parallel inference for large language models on resource-constrained devices.Proceedings of Machine Learning and Systems, 6:162–172, 2024

    2024

  10. [10]

    Powerinfer: Fast large language model serving with a consumer-grade gpu

    Yixin Song, Zeyu Mi, Haotong Xie, and Haibo Chen. Powerinfer: Fast large language model serving with a consumer-grade gpu. InProceedings of the ACM SIGOPS 30th Symposium on Operating Systems Principles, pages 590–606, 2024

    2024

  11. [11]

    Aser: activation smoothing and error reconstruction for large language model quantization

    Weibo Zhao, Yubin Shi, Xinyu Lyu, Wanchen Sui, Shen Li, and Yong Li. Aser: activation smoothing and error reconstruction for large language model quantization. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 22822–22830, 2025

    2025

  12. [12]

    Eora: Fine-tuning-free compensation for compressed llm with eigenspace low-rank approximation

    Shih-Yang Liu, Maksim Khadkevich, Nai Chit Fung, Charbel Sakr, Chao-Han Huck Yang, Chien-Yi Wang, Saurav Muralidharan, Hongxu Yin, Kwang-Ting Cheng, Jan Kautz, et al. Eora: Fine-tuning-free compensation for compressed llm with eigenspace low-rank approximation. arXiv preprint arXiv:2410.21271, 2024

    work page arXiv 2024

  13. [13]

    The Llama 3 Herd of Models

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models.arXiv preprint arXiv:2407.21783, 2024

    work page internal anchor Pith review arXiv 2024

  14. [14]

    Abq-llm: Arbitrary-bit quantized inference acceleration for large language models

    Chao Zeng, Songwei Liu, Yusheng Xie, Hong Liu, Xiaojian Wang, Miao Wei, Shu Yang, Fangmin Chen, and Xing Mei. Abq-llm: Arbitrary-bit quantized inference acceleration for large language models. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 22299–22307, 2025

    2025

  15. [15]

    Tim Dettmers, Mike Lewis, Younes Belkada, and Luke Zettlemoyer. Gpt3. int8 (): 8-bit matrix multiplication for transformers at scale.Advances in neural information processing systems, 35:30318–30332, 2022. 10

    2022

  16. [16]

    Mobilequant: Mobile-friendly quantization for on-device language models.arXiv preprint arXiv:2408.13933, 2024

    Fuwen Tan, Royson Lee, Shell Xu Hu, Sourav Bhattacharya, Timothy Hospedales, Georgios Tzimiropoulos, Brais Martinez, et al. Mobilequant: Mobile-friendly quantization for on-device language models.arXiv preprint arXiv:2408.13933, 2024

    work page arXiv 2024

  17. [17]

    Decdec: A systems approach to advancing low-bit llm quantization

    Yeonhong Park, Jake Hyun, Hojoon Kim, and Jae W Lee. Decdec: A systems approach to advancing low-bit llm quantization. In19th USENIX Symposium on Operating Systems Design and Implementation (OSDI 25), pages 803–819, 2025

    2025

  18. [18]

    Dl-qat: Weight-decomposed low- rank quantization-aware training for large language models.arXiv preprint arXiv:2504.09223, 2025

    Wenjin Ke, Zhe Li, Dong Li, Lu Tian, and Emad Barsoum. Dl-qat: Weight-decomposed low- rank quantization-aware training for large language models.arXiv preprint arXiv:2504.09223, 2025

    work page arXiv 2025

  19. [19]

    LLM-QAT: Data-free quantization aware training for large language models.arXiv:2305.17888, 2023

    Zechun Liu, Barlas Oguz, Changsheng Zhao, Ernie Chang, Pierre Stock, Yashar Mehdad, Yangyang Shi, Raghuraman Krishnamoorthi, and Vikas Chandra. Llm-qat: Data-free quantiza- tion aware training for large language models.arXiv preprint arXiv:2305.17888, 2023

    work page arXiv 2023

  20. [20]

    Accelerate: Training and inference at scale made simple, efficient and adaptable.https://github.com/huggingface/accelerate, 2022

    Sylvain Gugger, Lysandre Debut, Thomas Wolf, Philipp Schmid, Zachary Mueller, Sourab Mangrulkar, Marc Sun, and Benjamin Bossan. Accelerate: Training and inference at scale made simple, efficient and adaptable.https://github.com/huggingface/accelerate, 2022

    2022

  21. [21]

    Deepspeed-inference: enabling efficient inference of transformer models at unprecedented scale

    Reza Yazdani Aminabadi, Samyam Rajbhandari, Ammar Ahmad Awan, Cheng Li, Du Li, et al. Deepspeed-inference: enabling efficient inference of transformer models at unprecedented scale. InSC22: International Conference for High Performance Computing, Networking, Storage and Analysis, pages 1–15. IEEE, 2022

    2022

  22. [22]

    Fiddler: Cpu-gpu orchestration for fast inference of mixture-of-experts models.arXiv preprint arXiv:2402.07033, 2024

    Keisuke Kamahori, Tian Tang, Yile Gu, Kan Zhu, and Baris Kasikci. Fiddler: Cpu-gpu orchestration for fast inference of mixture-of-experts models.arXiv preprint arXiv:2402.07033, 2024

    work page arXiv 2024

  23. [23]

    Hobbit: A mixed precision expert offloading system for fast moe inference.arXiv preprint arXiv:2411.01433, 2024

    Peng Tang, Jiacheng Liu, Xiaofeng Hou, Yifei Pu, Jing Wang, Pheng-Ann Heng, Chao Li, and Minyi Guo. Hobbit: A mixed precision expert offloading system for fast moe inference.arXiv preprint arXiv:2411.01433, 2024

    work page arXiv 2024

  24. [24]

    Lq-lora: Low-rank plus quantized matrix decomposition for efficient language model finetuning,

    Han Guo, Philip Greengard, Eric P Xing, and Yoon Kim. Lq-lora: Low-rank plus quantized matrix decomposition for efficient language model finetuning.arXiv preprint arXiv:2311.12023, 2023

    work page arXiv 2023

  25. [25]

    Loftq: Lora-fine-tuning-aware quantization for large language models,

    Yixiao Li, Yifan Yu, Chen Liang, Pengcheng He, Nikos Karampatziakis, Weizhu Chen, and Tuo Zhao. Loftq: Lora-fine-tuning-aware quantization for large language models.arXiv preprint arXiv:2310.08659, 2023

    work page arXiv 2023

  26. [26]

    Lorc: Low-rank compression for llms kv cache with a progressive compression strategy.arXiv preprint arXiv:2410.03111,

    Rongzhi Zhang, Kuang Wang, Liyuan Liu, Shuohang Wang, Hao Cheng, Chao Zhang, and Yelong Shen. Lorc: Low-rank compression for llms kv cache with a progressive compression strategy.arXiv preprint arXiv:2410.03111, 2024

    work page arXiv 2024

  27. [27]

    Transformers: State- of-the-art natural language processing

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, et al. Transformers: State- of-the-art natural language processing. InProceedings of the 2020 conference on empirical methods in natural language processing: system demonstrations, pages 38–45, 2020

    2020

  28. [28]

    Gerganov and contributors

    G. Gerganov and contributors. llama.cpp: Inference of llama model in pure c/c++. https: //github.com/ggml-org/llama.cpp, 2025

    2025

  29. [29]

    Efficient memory management for large language model serving with pagedattention

    Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, and Ion Stoica. Efficient memory management for large language model serving with pagedattention. InProceedings of the 29th symposium on operating systems principles, pages 611–626, 2023

    2023

  30. [30]

    Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge

    Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge.arXiv preprint arXiv:1803.05457, 2018. 11

    work page internal anchor Pith review arXiv 2018

  31. [31]

    Mathqa: Towards interpretable math word problem solving with operation-based formalisms

    Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh Hajishirzi. Mathqa: Towards interpretable math word problem solving with operation-based formalisms. InProceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and ...

    2019

  32. [32]

    Measuring Massive Multitask Language Understanding

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding.arXiv preprint arXiv:2009.03300, 2020

    work page internal anchor Pith review arXiv 2009

  33. [33]

    Documenting large webtext corpora: A case study on the colossal clean crawled corpus

    Jesse Dodge, Maarten Sap, Ana Marasovi´c, William Agnew, Gabriel Ilharco, Dirk Groeneveld, Margaret Mitchell, and Matt Gardner. Documenting large webtext corpora: A case study on the colossal clean crawled corpus. InProceedings of the 2021 conference on empirical methods in natural language processing, pages 1286–1305, 2021. 12 A Implementation Details an...

    2021

  34. [34]

    Writes the current activation tensor into shared memory

  35. [35]

    Signals the CPU process via an event to start compensation computation

  36. [36]

    Immediately continues executing the quantized backbone operators. After completing the corresponding backbone computation, the GPU process waits for the CPU process to finish, retrieves the compensated outputs from shared memory, and merges them with the current activations before proceeding. From the CPU process perspective, it acts as a lightweight task...

  37. [37]

    Reads the activation tensor from shared memory

  38. [38]

    Determines the current execution context based on its locally maintained state

  39. [39]

    Selects the corresponding subset of LoRA parameters according to the dynamic rank allocation results

  40. [40]

    Computes the compensation outputs sequentially

  41. [41]

    Limitations

    Writes the results back to shared memory and signals completion. To reduce communication overhead, the CPU process maintains its own synchronized execution state (e.g., current layer and projection index), eliminating the need for the GPU process to repeatedly transmit meta-information. As a result, only activation tensors are exchanged during runtime. A....

  42. [42]

    Guidelines: • The answer [N/A] means that the paper does not involve crowdsourcing nor research with human subjects

    Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals (or an equivalent approval/review based on the requirements of your country or ...