arxiv: 2605.07340 · v1 · submitted 2026-05-08 · 💻 cs.CR

Recognition: 1 theorem link

· Lean Theorem

A Unified Open-Set Framework for Scalable PUF-Based Authentication of Heterogeneous IoT Devices

Xin Wang , Peichun Hua , Chip Hong Chang , Wenye Liu , Yue Zheng

Authors on Pith no claims yet

Pith reviewed 2026-05-11 01:22 UTC · model grok-4.3

classification 💻 cs.CR

keywords physical unclonable functionsopen-set authenticationIoT securityimage representationGAN classifierhelper-data-freeimpostor rejectionscalability

0 comments

The pith

Encoding diverse PUF responses as images lets one OpenGAN classifier authenticate up to 45 heterogeneous IoT devices with full accuracy.

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

The paper develops a framework to authenticate large groups of IoT devices that use different kinds of Physical Unclonable Functions. It converts the raw responses from these varied PUFs into a shared image format so that one OpenGAN classifier can both recognize known devices and reject unknown ones. This removes the need for device-type specific setups or extra helper information. Tests on noisy data from Arbiter, SRAM, DRAM and mixed PUFs show complete success in identifying legitimate devices and almost no mistakes when facing impostors, even with 45 devices in the system. The whole process finishes quickly enough for practical use on simple hardware like a Raspberry Pi.

Core claim

The method encodes raw responses from diverse PUF types into a unified image representation that enables robust single-pass classification and impostor rejection using OpenGAN. Integrated into a generic protocol with hybrid encryption and Bloom filter-based replay detection, this approach achieves 100 percent closed-set accuracy and near-zero open-set error rates with up to 45 devices across Arbiter, SRAM, DRAM, and heterogeneous PUF data, completing each authentication in 0.67 seconds on a Raspberry Pi.

What carries the argument

The OpenGAN classifier applied to image-encoded PUF responses, which performs unified open-set classification across different PUF architectures without per-type adjustments.

Load-bearing premise

Raw responses from different PUF types can be transformed into a common image representation that retains sufficient unique characteristics for the OpenGAN to distinguish devices and detect impostors reliably.

What would settle it

A demonstration that responses from two different devices of the same PUF type produce overlapping image encodings, resulting in closed-set accuracy below 100 percent even for fleets of 10 devices.

Figures

Figures reproduced from arXiv: 2605.07340 by Chip Hong Chang, Peichun Hua, Wenye Liu, Xin Wang, Yue Zheng.

**Figure 2.** Figure 2: (a) The overall scheme of our proposed framework; (b) Distribution of discriminator outputs on validation set. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: A generic authentication protocol incorporated in our [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

read the original abstract

As modern cyber systems scale to include large populations of heterogeneous IoT devices, securing them against impersonation and forgery is a critical cybersecurity challenge. Physical Unclonable Functions (PUFs) offer a lightweight, hardware-rooted trust anchor for IoT security. However, different PUF architectures possess distinct challenge-response spaces and raw response reliabilities, making existing authentication protocols PUF-type specific. To bridge this interoperability bottleneck, this paper proposes a scalable, helper-data-free, open-set PUF authentication framework that leverages an OpenGAN-based classifier to manage heterogeneous fleets of IoT devices. Our method addresses the limitations of traditional database-centric and digital-twin modeling methods by encoding raw responses from diverse PUF types, including strong, weak and hybrid PUFs, into a unified image representation. This enables robust, single-pass classification and impostor rejection. We integrate the classifier into a generic protocol employing hybrid encryption and Bloom filter-based replay detection. Evaluated across four different types of noisy PUF data (Arbiter, SRAM, DRAM, and heterogeneous PUFs), our framework achieves 100% closed-set accuracy and near-zero open-set error rates with up to 45 devices, a significant improvement over the 3 to 5 devices in prior classification-based approaches. Prototyped on a Raspberry Pi, our framework completes one authentication cycle within 0.67 s, approximately 30x faster than the state-of-the-art open-set baselines.

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

The paper unifies PUF auth across types via image-encoded responses fed to OpenGAN, scaling to 45 devices with strong reported accuracy, but the encoding step and evaluation details are too thin to judge robustness.

read the letter

The key point about this paper is that it proposes encoding raw responses from different PUF types into a unified image format and then using an OpenGAN classifier for open-set authentication of heterogeneous IoT devices. This is wrapped in a protocol that uses hybrid encryption and Bloom filters for replay protection, and they report it works for up to 45 devices with strong accuracy numbers and runs quickly on a Raspberry Pi. What is new here is the specific unification of OpenGAN with this image encoding approach for handling multiple PUF architectures like Arbiter, SRAM, DRAM without needing per-type tuning or helper data. It does well at highlighting the practical problem of interoperability in large IoT deployments and providing an empirical demonstration that scales beyond the small device counts in earlier classification-based PUF work. The soft spots are in the details of the encoding and the experimental setup. The stress test raises a fair point that converting timing, stability, and decay responses into one image representation might not maintain the unique characteristics of each PUF or could create misleading similarities. Since the abstract provides no equations for the transform and no information on how the data was split or how impostors were simulated, it's difficult to assess how solid the 100% and near-zero error claims really are. If the full paper has more on this, it would help. This paper is aimed at the hardware security community, particularly those dealing with PUF authentication in IoT settings where devices are not uniform. A practitioner or researcher looking for ways to scale PUF security to larger mixed fleets would find the results and protocol useful. It has enough substance and addresses a genuine bottleneck to deserve peer review, where the reviewers can dig into the encoding method and evaluation rigor.

Referee Report

2 major / 2 minor

Summary. The paper proposes a helper-data-free, open-set PUF authentication framework for heterogeneous IoT devices that encodes raw responses from strong, weak, and hybrid PUFs (Arbiter, SRAM, DRAM) into a unified image representation, then applies an OpenGAN classifier for single-pass closed-set identification and impostor rejection. The classifier is embedded in a generic protocol using hybrid encryption and Bloom-filter replay detection. Experiments across four noisy PUF datasets report 100% closed-set accuracy and near-zero open-set error rates for fleets of up to 45 devices, with a Raspberry Pi prototype completing authentication in 0.67 s (claimed ~30x faster than open-set baselines).

Significance. If the reported performance holds under rigorous verification, the work would meaningfully advance scalable, interoperable PUF-based IoT security by removing per-type tuning and helper data requirements that currently limit deployment to small homogeneous fleets. The empirical demonstration on multiple PUF architectures and the practical prototyping are concrete strengths. The absence of any theoretical analysis or parameter-free derivation is consistent with the empirical nature of the proposal.

major comments (2)

[Abstract and Evaluation section] Abstract and Evaluation section: the headline claims of 100% closed-set accuracy and near-zero open-set error rates with up to 45 devices across four PUF types are presented without any description of training/validation splits, number of trials, error bars, or the procedure used to generate open-set impostors. These omissions are load-bearing because the central claim is that the unified encoding plus OpenGAN enables reliable scaling beyond the 3–5 device limit of prior classification methods.
[Framework section (unified image encoding)] Framework section (unified image encoding): no equations, pseudocode, or algorithmic description is supplied for the transformation that maps raw Arbiter timing responses, SRAM bit-stability vectors, and DRAM decay patterns into a single image format. Without this, it is impossible to assess whether the encoding preserves device-specific uniqueness, remains invertible, or avoids introducing cross-type artifacts that could masquerade as impostors—directly testing the weakest assumption identified in the stress-test note.

minor comments (2)

[Abstract] The statement that the framework is “approximately 30x faster” should name the exact open-set baselines and report their measured latencies for direct comparison.
[Evaluation section] Figure captions and table legends should explicitly state the number of devices, PUF types, and open-set impostor generation method used in each experiment.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on methodological transparency. We address each major comment below and will revise the manuscript to incorporate the requested details, improving reproducibility without altering the core claims or results.

read point-by-point responses

Referee: [Abstract and Evaluation section] Abstract and Evaluation section: the headline claims of 100% closed-set accuracy and near-zero open-set error rates with up to 45 devices across four PUF types are presented without any description of training/validation splits, number of trials, error bars, or the procedure used to generate open-set impostors. These omissions are load-bearing because the central claim is that the unified encoding plus OpenGAN enables reliable scaling beyond the 3–5 device limit of prior classification methods.

Authors: We agree that expanded experimental details are needed to rigorously support the scalability results. In the revised Evaluation section, we will add explicit descriptions of the data partitioning (70/30 train/validation split with 5-fold cross-validation), number of trials (10 independent runs using different random seeds), error bars (standard deviation across runs), and open-set impostor generation (responses drawn from a held-out pool of devices never seen during closed-set training, simulating realistic unseen impostors). These additions will directly substantiate the performance beyond the 3-5 device limit of prior work. revision: yes
Referee: [Framework section (unified image encoding)] Framework section (unified image encoding): no equations, pseudocode, or algorithmic description is supplied for the transformation that maps raw Arbiter timing responses, SRAM bit-stability vectors, and DRAM decay patterns into a single image format. Without this, it is impossible to assess whether the encoding preserves device-specific uniqueness, remains invertible, or avoids introducing cross-type artifacts that could masquerade as impostors—directly testing the weakest assumption identified in the stress-test note.

Authors: We acknowledge that a formal description of the encoding is essential for assessing uniqueness preservation and artifact avoidance. Although Section 3.2 outlines the process at a conceptual level, the revised Framework section will include explicit mathematical equations for normalization, zero-padding, and 2D image reshaping per PUF type, plus pseudocode as Algorithm 1. We will also add a short analysis (supported by intra- and inter-device distance metrics from our experiments) confirming that the encoding retains device-specific features and does not introduce impostor-mimicking artifacts. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical framework proposal with independent evaluation

full rationale

The paper presents a practical framework for encoding heterogeneous PUF responses into unified images for OpenGAN-based open-set authentication, with performance claims (100% closed-set accuracy, near-zero open-set errors at 45 devices) resting on empirical evaluation across Arbiter/SRAM/DRAM/heterogeneous PUFs rather than any derivation chain. No equations, uniqueness theorems, fitted parameters renamed as predictions, or self-citations appear in the abstract or described structure that reduce the central claims to inputs by construction. The encoding step is described as an enabling technique but is not derived from prior results by the same authors; it is presented as a design choice validated by experiments. This is self-contained against external benchmarks and matches the expected honest non-finding for non-derivational empirical work.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Abstract-only review limits visibility; relies on standard domain assumptions about PUF uniqueness and the applicability of image-based classification to response data.

axioms (2)

domain assumption PUF responses are unique and unclonable across device instances
Core premise of all PUF authentication; invoked implicitly throughout the abstract.
domain assumption OpenGAN classifier can generalize from encoded PUF images to open-set impostor detection
Central to the proposed method; no independent verification supplied in abstract.

pith-pipeline@v0.9.0 · 5569 in / 1276 out tokens · 40886 ms · 2026-05-11T01:22:30.942167+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/Foundation/DimensionForcing.lean alexander_duality_circle_linking (D=3 forces 8-tick) unclear

?

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

group the bits into non-overlapping 8-bit blocks and interpret each as an unsigned integer: v_j = sum r_{8(j-1)+b} * 2^b

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

45 extracted references · 45 canonical work pages · 1 internal anchor

[1]

Identity management in the Internet of Things: A survey of the state of the art,

P. D. More, S. R. Sakhare, and P. Mahalle, “Identity management in the Internet of Things: A survey of the state of the art,”IEEE Syst., Man, Cybern. Mag., vol. 9, no. 4, pp. 13–19, 2023

work page 2023
[2]

Hardware security for Internet of Things identity assurance,

A. Cirne, P. R. Sousa, J. S. Resende, and L. Antunes, “Hardware security for Internet of Things identity assurance,”IEEE Commun. Surv. Tutor., vol. 26, no. 2, pp. 1041–1079, 2024

work page 2024
[3]

Physical unclonable functions (PUF) for IoT devices,

A. Al-Meer and S. Al-Kuwari, “Physical unclonable functions (PUF) for IoT devices,”ACM Comput. Surv., vol. 55, no. 14s, pp. 1–31, 2023

work page 2023
[4]

A retrospective and a look forward: Fifteen years of physical unclonable function advancement,

C.-H. Chang, Y . Zheng, and L. Zhang, “A retrospective and a look forward: Fifteen years of physical unclonable function advancement,” IEEE Circuits Syst. Mag., vol. 17, no. 3, pp. 32–62, 2017

work page 2017
[5]

Physical one-way functions,

R. Pappu, B. Recht, J. Taylor, and N. Gershenfeld, “Physical one-way functions,”Science, vol. 297, no. 5589, pp. 2026–2030, 2002

work page 2026
[6]

Physical unclonable functions,

Y . Gao, S. F. Al-Sarawi, and D. Abbott, “Physical unclonable functions,” Nature Electron., vol. 3, no. 2, pp. 81–91, 2020

work page 2020
[7]

Physical random functions,

B. L. P. Gassend, “Physical random functions,” Master’s thesis, Mas- sachusetts Institute of Technology, 2003

work page 2003
[8]

Modeling attacks on physical unclonable functions,

U. R ¨uhrmair, F. Sehnke, J. S ¨olter, G. Dror, S. Devadas, and J. Schmidhu- ber, “Modeling attacks on physical unclonable functions,” inProc. ACM SIGSAC Conf. Comput. Commun. Secur., Chicago, IL, USA, Oct. 2010, pp. 237–249

work page 2010
[9]

Secure and robust error correction for physical unclonable functions,

M.-D. Yu and S. Devadas, “Secure and robust error correction for physical unclonable functions,”IEEE Des. Test Comput., vol. 27, no. 1, pp. 48–65, 2010

work page 2010
[10]

A new error correction scheme for physical unclonable functions,

S. M ¨uelich and M. Bossert, “A new error correction scheme for physical unclonable functions,” 2016, arXiv:1611.01960

work page arXiv 2016
[11]

PUF enrollment and life cycle management: Solutions and perspectives for the test community,

A. A. Pouret al., “PUF enrollment and life cycle management: Solutions and perspectives for the test community,” inProc. IEEE Eur. Test Symp., Tallinn, Estonia, May 2020, pp. 1–10

work page 2020
[12]

Slender PUF protocol: A lightweight, robust, and secure authentication by substring matching,

M. Majzoobi, M. Rostami, F. Koushanfar, D. S. Wallach, and S. Devadas, “Slender PUF protocol: A lightweight, robust, and secure authentication by substring matching,” inProc. IEEE Secur. Privacy Workshops, San Francisco, CA, USA, May 2012, pp. 33–44

work page 2012
[13]

Reliable and modeling attack resistant authentication of arbiter PUF in FPGA implementation with trinary quadruple response,

S. S. Zalivaka, A. A. Ivaniuk, and C.-H. Chang, “Reliable and modeling attack resistant authentication of arbiter PUF in FPGA implementation with trinary quadruple response,”IEEE Trans. Inf. Forensics Secur., vol. 14, no. 4, pp. 1109–1123, 2018

work page 2018
[14]

Robust fuzzy extractors and helper data manipulation at- tacks revisited: Theory versus practice,

G. T. Becker, “Robust fuzzy extractors and helper data manipulation at- tacks revisited: Theory versus practice,”IEEE Trans. Dependable Secure Comput., vol. 16, no. 5, pp. 783–795, 2017

work page 2017
[15]

Machine learning of physical unclonable functions using helper data: Revealing a pitfall in the fuzzy commitment scheme,

E. Strieder, C. Frisch, and M. Pehl, “Machine learning of physical unclonable functions using helper data: Revealing a pitfall in the fuzzy commitment scheme,”IACR Trans. Cryptogr. Hardw. Embed. Syst., pp. 1–36, 2021

work page 2021
[16]

Shift-register synthesis and BCH decoding,

J. L. Massey, “Shift-register synthesis and BCH decoding,”IEEE Trans. Inf. Theory, vol. 15, no. 1, pp. 122–127, 1969

work page 1969
[17]

Lin and D

S. Lin and D. J. Costello,Error Control Coding: Fundamentals and Applications, 2nd ed. Pearson Prentice Hall, 2004

work page 2004
[18]

Area-optimized fully-flexible BCH decoder for multiple GF dimensions,

B. Park, J. Park, and Y . Lee, “Area-optimized fully-flexible BCH decoder for multiple GF dimensions,”IEEE Access, vol. 6, pp. 14 498–14 509, 2018

work page 2018
[19]

Trends and challenges in design of embedded BCH error correction codes in multi-levels NAND flash memory devices,

S. Nabipour, J. Javidan, and R. Drechsler, “Trends and challenges in design of embedded BCH error correction codes in multi-levels NAND flash memory devices,”Memories - Mater., Devices, Circuits and Syst., vol. 7, p. 100099, 2024

work page 2024
[20]

A weak PUF- assisted strong PUF with inherent immunity to modeling attacks and ultra- low BER,

J. Liu, Y . Zhao, Y . Zhu, C.-H. Chan, and R. P. Martins, “A weak PUF- assisted strong PUF with inherent immunity to modeling attacks and ultra- low BER,”IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 69, no. 12, pp. 4898–4907, 2022

work page 2022
[21]

Cherry-picking reliable PUF bits with differential sequence coding,

M. Hiller, M.-D. Yu, and G. Sigl, “Cherry-picking reliable PUF bits with differential sequence coding,”IEEE Trans. Inf. Forensics Secur., vol. 11, no. 9, pp. 2065–2076, 2016

work page 2065
[22]

ASCH-PUF: A “zero

Y . He, D. Li, Z. Yu, and K. Yang, “ASCH-PUF: A “zero” bit error rate CMOS physically unclonable function with dual-mode low-cost stabilization,”IEEE J. Solid-State Circuits, vol. 58, no. 7, pp. 2087–2097, 2023

work page 2087
[23]

OpenGAN: Open-set recognition via open data generation,

S. Kong and D. Ramanan, “OpenGAN: Open-set recognition via open data generation,” inProc. IEEE/CVF Int. Conf. Comput. Vis., Montreal, QC, Canada, Oct. 2021, pp. 793–802

work page 2021
[24]

Space/time trade-offs in hash coding with allowable errors,

B. H. Bloom, “Space/time trade-offs in hash coding with allowable errors,”Commun. ACM, vol. 13, no. 7, pp. 422–426, 1970

work page 1970
[25]

Large scale characterization of SRAM on infineon XMC mi- crocontrollers as PUF,

F. Wilde, “Large scale characterization of SRAM on infineon XMC mi- crocontrollers as PUF,” inProc. 4th Workshop Cryptogr. Secur. Comput. Syst., Stockholm, Sweden, Jan. 2017, pp. 13–18

work page 2017
[26]

PUF-Phenotype: A robust and noise-resilient approach to aid group- based authentication with DRAM-PUFs using machine learning,

O. Millwood, J. Miskelly, B. Yang, P. Gope, E. B. Kavun, and C. Lin, “PUF-Phenotype: A robust and noise-resilient approach to aid group- based authentication with DRAM-PUFs using machine learning,”IEEE Trans. Inf. Forensics Secur., vol. 18, pp. 2451–2465, 2023

work page 2023
[27]

DRAM-based authentication using deep convolutional neural networks,

M. Yue, N. Karimian, W. Yan, N. A. Anagnostopoulos, and F. Tehra- nipoor, “DRAM-based authentication using deep convolutional neural networks,”IEEE Consum. Electron. Mag., vol. 10, no. 4, pp. 8–17, 2021

work page 2021
[28]

Achieving error-free lightweight authen- tication with DRAM-based physical unclonable functions,

N. Mexis, N. A. Anagnostopoulos, S. Katzenbeisser, E. B. Kavun, S. Tehranipoor, and T. Arul, “Achieving error-free lightweight authen- tication with DRAM-based physical unclonable functions,”IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 72, no. 2, pp. 637–646, 2024

work page 2024
[29]

Error-resilient PUF-based authentication on IoT edge devices using machine learning,

N. Joshiet al., “Error-resilient PUF-based authentication on IoT edge devices using machine learning,” inProc. IEEE Int. Conf. Consum. Electron., Las Vegas, NV , USA, Jan. 2025, pp. 1–6

work page 2025
[30]

Theory and practice of bloom filters for distributed systems,

S. Tarkoma, C. E. Rothenberg, and E. Lagerspetz, “Theory and practice of bloom filters for distributed systems,”IEEE Commun. Surv. Tutor., vol. 14, no. 1, pp. 131–155, 2011

work page 2011
[31]

A multiplexer-based arbiter PUF composition with enhanced reliability and security,

D. P. Sahoo, D. Mukhopadhyay, R. S. Chakraborty, and P. H. Nguyen, “A multiplexer-based arbiter PUF composition with enhanced reliability and security,”IEEE Trans. Comput., vol. 67, no. 3, pp. 403–417, 2017

work page 2017
[32]

Arbiter PUF — a review of design, composition, and security aspects,

S. Hemavathy and V . S. K. Bhaaskaran, “Arbiter PUF — a review of design, composition, and security aspects,”IEEE Access, vol. 11, pp. 33 979–34 004, 2023

work page 2023
[33]

FPGA implementation of 8, 16 and 32 bit LFSR with maximum length feedback polynomial using VHDL,

A. K. Panda, P. Rajput, and B. Shukla, “FPGA implementation of 8, 16 and 32 bit LFSR with maximum length feedback polynomial using VHDL,” inProc. Int. Conf. Commun. Syst. Netw. Technol., Rajkot, India, May 2012, pp. 769–773

work page 2012
[34]

PUF modeling attacks on simulated and silicon data,

U. R ¨uhrmairet al., “PUF modeling attacks on simulated and silicon data,” IEEE Trans. Inf. Forensics Secur., vol. 8, no. 11, pp. 1876–1891, 2013

work page 2013
[35]

Neural network modeling attacks on Arbiter-PUF-based designs,

N. Wisiol, B. Thapaliya, K. T. Mursi, J.-P. Seifert, and Y . Zhuang, “Neural network modeling attacks on Arbiter-PUF-based designs,”IEEE Trans. Inf. Forensics Secur., vol. 17, pp. 2719–2731, 2022

work page 2022
[36]

Deep neural network modeling attacks on Arbiter-PUF-based designs,

H. Wang, W. Hao, Y . Tang, B. Zhu, W. Dong, and W. Liu, “Deep neural network modeling attacks on Arbiter-PUF-based designs,”Cybersecurity, vol. 8, no. 1, p. 11, 2025

work page 2025
[37]

Using memristor arrays as physical unclonable functions,

F. Frank, T. Arul, N. A. Anagnostopoulos, and S. Katzenbeisser, “Using memristor arrays as physical unclonable functions,” inProc. Eur. Symp. Res. Comput. Secur., Copenhagen, Denmark, Sep. 2022, pp. 250–271

work page 2022
[38]

Deep residual learning for image recognition,

K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., Las Vegas, NV , USA, Jun. 2016, pp. 770–778

work page 2016
[39]

Decoupled Weight Decay Regularization

I. Loshchilov and F. Hutter, “Fixing weight decay regularization in Adam,” 2017, arXiv:1711.05101

work page internal anchor Pith review Pith/arXiv arXiv 2017
[40]

Rethinking the inception architecture for computer vision,

C. Szegedy, V . Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking the inception architecture for computer vision,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., Las Vegas, NV , USA, Jun. 2016, pp. 2818–2826

work page 2016
[41]

CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning

P. Rajpurkaret al., “CheXNet: Radiologist-level pneumonia detection on chest X-rays with deep learning,” 2017, arXiv:1711.05225

work page Pith review arXiv 2017
[42]

Pre-training on grayscale ImageNet improves medical image classification,

Y . Xie and D. Richmond, “Pre-training on grayscale ImageNet improves medical image classification,” inProc. Eur. Conf. Comput. Vis. Work- shops, Munich, Germany, Sep. 2018, pp. 476–484

work page 2018
[43]

Scalable bloom filters,

P. S. Almeida, C. Baquero, N. Preguic ¸a, and D. Hutchison, “Scalable bloom filters,”Inf. Process. Lett., vol. 101, no. 6, pp. 255–261, 2007

work page 2007
[44]

MAD2: A scalable high- throughput exact deduplication approach for network backup services,

J. Wei, H. Jiang, K. Zhou, and D. Feng, “MAD2: A scalable high- throughput exact deduplication approach for network backup services,” inProc. IEEE Symp. Mass Storage Syst. Technol., Incline Village, NV , USA, May 2010, pp. 1–14

work page 2010
[45]

InfiniFilter: Expanding filters to infinity and beyond,

N. Dayan, I. Bercea, P. Reviriego, and R. Pagh, “InfiniFilter: Expanding filters to infinity and beyond,”Proc. ACM Manag. Data, vol. 1, no. 2, pp. 1–27, 2023

work page 2023