pith. sign in

arxiv: 2606.30584 · v1 · pith:6YBSJAU3new · submitted 2026-06-29 · 💻 cs.IT · math.IT

Semantic Noise Aided Secure Image Transmission over MIMO Fading Channels

Xue Han , Biqian Feng , Ting Zhou , Yongpeng Wu , Yuanwei Liu , Arumugam Nallanathan , Xiang-Gen Xia , Wenjun Zhang This is my paper

Pith reviewed 2026-06-30 03:16 UTC · model grok-4.3

classification 💻 cs.IT math.IT
keywords semantic communicationsecure image transmissionMIMO fading channelssemantic noisebeamforming optimizationchannel estimationeavesdropper interference
0
0 comments X

The pith

A semantic noise generation network interferes with an eavesdropper's image reconstruction while preserving high fidelity at the legitimate user over MIMO fading channels.

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

The paper introduces a secure image semantic communication framework that adds a tailored noise map to transmitted features. The noise is produced by a dedicated network that uses both the original image features and the legitimate receiver's channel state information. This selective interference is combined with beamforming optimization and improved channel estimation to maintain reconstruction quality for the intended user. If the approach holds, semantic communication systems could achieve security through noise design rather than separate encryption layers.

Core claim

The paper claims that generating semantic noise from source features and the legitimate user's CSI, paired with transceiver beamformer optimization solved via constrained stochastic successive convex approximation, enables secure image transmission over MIMO fading channels. Numerical results show that this combination protects image content from the eavesdropper while delivering high-fidelity reconstruction at the legitimate semantic user.

What carries the argument

The semantic noise generation network, which produces a noise map from source features and legitimate-user CSI to create selective interference.

If this is right

  • Semantic noise and beamforming together block information leakage to the eavesdropper.
  • The channel estimation network improves CSI accuracy and overall system performance.
  • The optimization algorithm yields practical beamformer solutions under the stated constraints.
  • High-fidelity reconstruction at the legitimate user is preserved alongside the security effect.

Where Pith is reading between the lines

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

  • The same noise-generation idea might extend to video or other semantic tasks if the network generalizes across data types.
  • Real deployments would need to verify performance when channel estimates contain errors.
  • The framework could reduce reliance on conventional cryptographic overhead in wireless image links.

Load-bearing premise

A noise map created using the legitimate user's channel information will disrupt the eavesdropper's reconstruction without lowering quality at the legitimate receiver.

What would settle it

A test in which the eavesdropper reconstructs the image with quality comparable to the legitimate user after the semantic noise is added.

Figures

Figures reproduced from arXiv: 2606.30584 by Arumugam Nallanathan, Biqian Feng, Ting Zhou, Wenjun Zhang, Xiang-Gen Xia, Xue Han, Yongpeng Wu, Yuanwei Liu.

Figure 1
Figure 1. Figure 1: (a) System model of semantic image communication network. (b) The [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) The detailed structure of the SISC semantic encoder network. (b) Two consecutive Swin Transformer blocks. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The SN-MSA network and the detailed architecture of the dual CSI [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The architecture of the channel estimation enhanced network (CEEN) and the enhanced depthwise separable convolution (EDSC) network. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) The semantic noise generation mechanism of the SNG network. (b) The detailed SN-CVAE structure. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison of different schemes versus transmit power on the CVRG-Pano dataset over MIMO fading channels. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Security performance comparison of different schemes for SU and Eve [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparisons of PSNR improvement of varying learnable semantic [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Visualization of reconstructed images obtained by different schemes. The values below the images represent the PSNR(dB)/MS-SSIM results. The first [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗
read the original abstract

Existing semantic communications have exhibited satisfactory performance in many tasks, but secure image transmission remains insufficiently explored. We propose a novel secure image semantic communication (SISC) framework over multiple-input multiple-output (MIMO) fading channels. To ensure high-quality image reconstruction for the legitimate semantic user (SU) and simultaneously interfere with the eavesdropper (Eve), we design a semantic noise generation (SNG) network. This network generates a beneficial semantic noise map based on both the source features and the SU channel state information (CSI). An efficient channel estimation enhanced network is incorporated to obtain the accurate CSI and enhance the system performance. Furthermore, to improve the secure image reconstruction quality, we develop an efficient transceiver beamformer optimization algorithm, where the formulated problem is solved using the constrained stochastic successive convex approximation method. In the proposed SISC framework, semantic noise generation and beamforming optimization work together to ensure secure and high-quality image transmission. Numerical results demonstrate that the proposed semantic noise aided transmission scheme effectively protects image information from leakage to Eve while maintaining high-fidelity image reconstruction at SU.

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

3 major / 2 minor

Summary. The paper proposes a secure image semantic communication (SISC) framework for MIMO fading channels. A semantic noise generation (SNG) network produces a noise map from source features and legitimate-user CSI to degrade eavesdropper reconstruction while preserving high-fidelity reconstruction at the semantic user (SU). An enhanced channel-estimation network is included, and transceiver beamformers are optimized via constrained stochastic successive convex approximation. Numerical results are presented to support that the scheme protects image information from leakage to Eve while maintaining SU fidelity.

Significance. If the selective-interference property of the SNG holds beyond the simulated channel realizations, the work would offer a novel integration of learned semantic noise with physical-layer beamforming for secure semantic communications. The combination of data-driven noise generation with stochastic convex approximation is a potentially useful direction, though the absence of analytic bounds or robustness guarantees limits the strength of the contribution.

major comments (3)
  1. [Framework description and SNG network section] The central security claim rests on the SNG producing a CSI-tailored noise map that exploits statistical differences between h_SU and h_Eve to create destructive interference at Eve while remaining neutral or constructive at SU. No analytic bound on the resulting mutual-information gap or reconstruction-metric difference is derived, and no analysis is given for correlated fading or imperfect CSI. This assumption is load-bearing for the claim that numerical results demonstrate general protection rather than simulation-specific behavior.
  2. [Beamformer optimization section] The beamforming optimization is solved with constrained stochastic successive convex approximation, yet the manuscript provides neither convergence guarantees nor an analysis of how estimation errors in the channel-estimation network propagate into the effective noise at Eve. Without these, it is unclear whether the reported security gains are robust or depend on idealized channel knowledge.
  3. [Numerical results section] Numerical results are invoked to support the selective-interference claim, but the evaluation does not report performance under channel correlation coefficients or CSI estimation error variances. If the training distribution does not sufficiently separate the two channels, the observed gap may not generalize.
minor comments (2)
  1. [System model] Notation for the SNG output (noise map) and its interaction with the precoder should be defined more explicitly to avoid ambiguity when describing the effective received signals at SU and Eve.
  2. [Numerical results] The abstract states that 'numerical results demonstrate' the claims; the results section should include error bars or statistics over multiple independent channel realizations to substantiate this.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thorough review and insightful comments on our manuscript. We address each of the major comments point by point below, providing clarifications and indicating revisions where appropriate.

read point-by-point responses

  1. Referee: [Framework description and SNG network section] The central security claim rests on the SNG producing a CSI-tailored noise map that exploits statistical differences between h_SU and h_Eve to create destructive interference at Eve while remaining neutral or constructive at SU. No analytic bound on the resulting mutual-information gap or reconstruction-metric difference is derived, and no analysis is given for correlated fading or imperfect CSI. This assumption is load-bearing for the claim that numerical results demonstrate general protection rather than simulation-specific behavior.

    Authors: We acknowledge that the absence of analytic bounds limits the generality of the claims. Deriving such bounds for the data-driven SNG network is challenging due to its non-linear and learned nature. The security is demonstrated through extensive simulations under the MIMO fading channel model. In the revised manuscript, we will add a dedicated subsection discussing the assumptions, including the impact of channel correlation and imperfect CSI, and note this as a direction for future work. Additionally, we will include new simulation results for correlated channels to strengthen the evaluation. revision: partial

  2. Referee: [Beamformer optimization section] The beamforming optimization is solved with constrained stochastic successive convex approximation, yet the manuscript provides neither convergence guarantees nor an analysis of how estimation errors in the channel-estimation network propagate into the effective noise at Eve. Without these, it is unclear whether the reported security gains are robust or depend on idealized channel knowledge.

    Authors: The constrained stochastic successive convex approximation method is adopted from established optimization literature, where convergence properties have been analyzed under suitable conditions. We will cite the relevant references more explicitly. For the propagation of estimation errors, a complete theoretical analysis is complex given the joint training of the networks. We will add numerical experiments in the revision to evaluate the sensitivity to CSI estimation errors and their effect on the security performance at Eve. revision: partial

  3. Referee: [Numerical results section] Numerical results are invoked to support the selective-interference claim, but the evaluation does not report performance under channel correlation coefficients or CSI estimation error variances. If the training distribution does not sufficiently separate the two channels, the observed gap may not generalize.

    Authors: We agree that additional evaluations would enhance the robustness assessment. In the revised version, we will extend the numerical results to include performance metrics for various channel correlation coefficients and different levels of CSI estimation error variances. This will provide evidence on the generalization of the selective-interference property. revision: yes

Circularity Check

0 steps flagged

No circularity: framework proposal validated by independent simulation benchmarks

full rationale

The paper presents a proposed SISC framework consisting of an SNG network, channel estimation network, and beamforming optimization solved via constrained stochastic successive convex approximation. The central claims rest on numerical results from simulations demonstrating selective interference and reconstruction quality. No equations or steps in the provided description reduce a claimed prediction or uniqueness result to a fitted parameter or self-citation by construction. The optimization method is a standard algorithmic technique, and performance is assessed against external simulation benchmarks rather than internal redefinitions. This is the common case of a self-contained engineering proposal.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The central claim depends on the trained neural networks and optimization algorithm performing as intended under standard channel assumptions. Many free parameters in the ML components are expected. Without full text, the ledger is estimated from abstract descriptions.

free parameters (2)
  • Parameters of the SNG network
    The semantic noise generation network is a neural network whose weights are trained/fitted to data.
  • Beamforming optimization parameters
    The constrained stochastic successive convex approximation likely involves parameters tuned for the problem.
axioms (2)
  • domain assumption Standard MIMO fading channel model
    The framework is built on typical assumptions for MIMO fading channels in wireless communications.
  • domain assumption Availability of SU CSI for noise generation
    The SNG uses SU CSI, assuming it can be obtained accurately via the channel estimation network.
invented entities (1)
  • Semantic noise map no independent evidence
    purpose: To provide beneficial interference to Eve while preserving SU reconstruction
    A new concept introduced in the framework without mentioned external evidence.

pith-pipeline@v0.9.1-grok · 5738 in / 1568 out tokens · 66885 ms · 2026-06-30T03:16:32.495600+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

43 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

    Toward wisdom-evolutionary and primitive-concise 6G: A new paradigm of semantic communication networks,

    P. Zhang, W. Xu, H. Gao, and K. Niu, “Toward wisdom-evolutionary and primitive-concise 6G: A new paradigm of semantic communication networks,”Engineering, vol. 8, pp. 60–73, Jan. 2022

    2022

  2. [2]

    SCSC: A novel standards-compatible semantic communication frame- work for image transmission,

    X. Han, Y . Wu, Z. Gao, B. Feng, Y . Shi, D. G ¨und¨uz, and W. Zhang, “SCSC: A novel standards-compatible semantic communication frame- work for image transmission,”IEEE Trans. Commun., vol. 73, no. 8, pp. 5682–5698, Aug. 2025

    2025

  3. [3]

    Enhancing image privacy in semantic communication over wiretap channels leveraging differential privacy,

    W. Chen, S. Tang, and Q. Yang, “Enhancing image privacy in semantic communication over wiretap channels leveraging differential privacy,” in Proc. IEEE Int. Workshop Mach. Learn. Signal Process. (MLSP), London, United Kingdom, Sept. 2024, pp. 1–6

    2024

  4. [4]

    A survey on wireless security: Technical challenges, recent advances, and future trends,

    Y . Zou, J. Zhu, X. Wang, and L. Hanzo, “A survey on wireless security: Technical challenges, recent advances, and future trends,”Proc. IEEE, vol. 104, no. 9, pp. 1727–1765, Sept. 2016

    2016

  5. [5]

    On the secrecy capacity of MIMO wiretap channels: Convex reformulation and efficient numerical methods,

    A. Mukherjee, B. Ottersten, and L.-N. Tran, “On the secrecy capacity of MIMO wiretap channels: Convex reformulation and efficient numerical methods,”IEEE Trans. Commun., vol. 69, no. 10, pp. 6865–6878, Oct. 2021

    2021

  6. [6]

    Physical layer security in near-field communications: What will be changed?

    Z. Zhang, Y . Liu, Z.-J. Wang, X. Mu, and J. Chen, “Physical layer security in near-field communications: What will be changed?” Feb. 2023, arXiv:2302.04189. [Online]. Available: http://arxiv.org/abs/2302.04189

    work page arXiv 2023

  7. [7]

    Exploiting multi-layer refracting RIS-assisted receiver for HAP-SWIPT networks,

    K. An, Y . Sun, Z. Lin, Y . Zhu, W. Ni, and N. Al-Dhahir, “Exploiting multi-layer refracting RIS-assisted receiver for HAP-SWIPT networks,” IEEE Wireless Commun., vol. 23, no. 10, pp. 12638–12657, Oct. 2024

    2024

  8. [8]

    Refracting RIS aided hybrid satellite-terrestrial relay networks: Joint beamforming design and optimization,

    Z. Lin, H. Niu, K. An, Y . Wang, G. Zheng, and S. Chatzinotas, “Refracting RIS aided hybrid satellite-terrestrial relay networks: Joint beamforming design and optimization,”IEEE Trans. Aerosp. Electron. Syst., vol. 58, no. 4, pp. 3717–3724, Aug. 2022

    2022

  9. [9]

    Self-powered absorptive reconfigurable intelligent surfaces for securing satellite-terrestrial integrated networks,

    L. Zhiet al., “Self-powered absorptive reconfigurable intelligent surfaces for securing satellite-terrestrial integrated networks,”China Commun., vol. 21, no. 9, pp. 276–291, Sept. 2024

    2024

  10. [10]

    Artificial-noise-aided space–time line code for enhancing physical layer security of multiuser MIMO downlink transmission,

    J. Choi, J. Joung, and Y . -S. Cho, “Artificial-noise-aided space–time line code for enhancing physical layer security of multiuser MIMO downlink transmission,”IEEE Syst. J., vol. 16, no. 1, pp. 1289–1300, Mar. 2022

    2022

  11. [11]

    Artificial noise assisted secure interference networks with wireless power transfer,

    N. Zhao, Y . Cao, F. R. Yu, Y . Chen, M. Jin, and V . C. M. Leung, “Artificial noise assisted secure interference networks with wireless power transfer,”IEEE Trans. V eh. Technol., vol. 67, no. 2, pp. 1087–1098, Feb. 2018

    2018

  12. [12]

    Green interference based symbiotic security in integrated satellite-terrestrial communica- tions,

    Z. Yin, N. Cheng, T. H. Luan, Y . Hui and W. Wang, “Green interference based symbiotic security in integrated satellite-terrestrial communica- tions,”IEEE Trans. Wireless Commun., vol. 21, no. 11, pp. 9962–9973, Nov. 2022

    2022

  13. [13]

    DT-assisted multi- point symbiotic security in space-air-ground integrated networks,

    Z. Yin, N. Cheng, T. H. Luan, Y . Song and W. Wang, “DT-assisted multi- point symbiotic security in space-air-ground integrated networks,”IEEE Trans. Inf. F orensics Secur ., vol. 18, pp. 5721–5734, 2023

    2023

  14. [14]

    Deep artificial noise: Deep learning-based precoding optimization for artificial noise scheme,

    S. Yun, J. M. Kang, I. M. Kim, and J. Ha, “Deep artificial noise: Deep learning-based precoding optimization for artificial noise scheme,”IEEE Trans. V eh. Technol., vol. 69, no. 3, pp. 3465–3469, Mar. 2020

    2020

  15. [15]

    Rethinking wireless communication security in semantic Internet of Things,

    H. Du, J. Wang, D. Niyato, J. Kang, Z. Xiong, and M. Guizani, “Rethinking wireless communication security in semantic Internet of Things,”IEEE Wireless Commun., vol. 30, no. 3, pp. 36—43, Jun. 2023

    2023

  16. [16]

    Secure semantic communications: Challenges, approaches, and opportunities,

    M. Shen, M. Shen, J. Wang, H. Du, D. Niyato, X. Tang, and J. Kang, “Secure semantic communications: Challenges, approaches, and opportunities,”IEEE Netw., vol. 38, no. 4, pp. 197–206, Jul. 2024

    2024

  17. [17]

    Robust semantic communications with masked VQ-V AE enabled codebook,

    Q. Hu, G. Zhang, Z. Qin, Y . Cai, G. Yu, and G. Y . Li, “Robust semantic communications with masked VQ-V AE enabled codebook,”IEEE Trans. Wireless Commun., vol. 22, no. 12, pp. 8707–8722, Dec. 2023

    2023

  18. [18]

    Privacy-preserving task-oriented semantic communications against model inversion attacks,

    Y . Wang, S. Guo, Y . Deng, H. Zhang, and Y . Fang, “Privacy-preserving task-oriented semantic communications against model inversion attacks,” IEEE Trans. Wireless Commun., vol. 23, no. 8, pp. 10150–10165, Aug. 2024

    2024

  19. [19]

    Deep joint source-channel and encryption coding: secure semantic communications,

    T. -Y . Tung and D. G ¨und¨uz, “Deep joint source-channel and encryption coding: secure semantic communications,” inProc. IEEE Int. Conf. Commun. (ICC), Rome, Italy, May 2023, pp. 5620–5625

    2023

  20. [20]

    Privacy-aware communi- cation over a wiretap channel with generative networks,

    E. Erdemir, P. L. Dragotti, and D. G ¨und¨uz, “Privacy-aware communi- cation over a wiretap channel with generative networks,” inIEEE Int. Conf. Acoust. Speech Signal Process. (ICASSP), Singapore, May 2022, pp. 2989–2993

    2022

  21. [21]

    Wireless image trans- mission with semantic and security awareness,

    M. Zhang, Y . Li, Z. Zhang, G. Zhu, and C. Zhong, “Wireless image trans- mission with semantic and security awareness,”IEEE Wireless Commun. Lett., vol. 12, no. 8, pp. 1389–1393, Aug. 2023

    2023

  22. [22]

    Secure semantic communications: From perspective of physical layer security,

    Y . Li, Z. Shi, H. Hu, Y . Fu, H. Wang, and H. Lei, “Secure semantic communications: From perspective of physical layer security,”IEEE Commun. Lett., vol. 28, no. 10, pp. 2243–2247, Oct. 2024

    2024

  23. [23]

    Privacy- preserving resource allocation for asynchronous federated learning,

    X. Chen, Z. Zhou, W. Ni, G. Pan, X. Wang, and S. Zhang, “Privacy- preserving resource allocation for asynchronous federated learning,” in Proc. IEEE 99th V eh. Technol. Conf. (VTC2024-Spring), Singapore, Jun. 2024, pp. 1–6

    2024

  24. [24]

    Encrypted semantic communi- cation using adversarial training for privacy preserving,

    X. Luo, Z. Chen, M. Tao, and F. Yang, “Encrypted semantic communi- cation using adversarial training for privacy preserving,”IEEE Commun. Lett., vol. 27, no. 6, pp. 1486–1490, Jun. 2023

    2023

  25. [25]

    SemProtector: A unified framework for semantic protection in deep learning-based semantic communication systems,

    X. Liu, G. Nan, Q. Cui, Z. Li, P. Liu, and Z. Xing, “SemProtector: A unified framework for semantic protection in deep learning-based semantic communication systems,”IEEE Commun. Mag., vol. 61, no. 11, pp. 56– 62, Nov. 2023

    2023

  26. [26]

    Semantic and technical noise modeling in semantic image communication over wireless channels,

    P. Samarathunga, N. Pollwaththage, T. Fernandoet al., “Semantic and technical noise modeling in semantic image communication over wireless channels,”IEEE Access, vol. 13, pp. 150872–150900, Aug. 2025

    2025

  27. [27]

    Robust semantic communications against semantic noise,

    Q. Hu, G. Zhang, Z. Qin, Y . Cai, G. Yu, and G. Y . Li, “Robust semantic communications against semantic noise,” inProc. IEEE 96th V eh. Technol. Conf. (VTC2022-Fall), London, United Kingdom, Sept. 2022, pp. 1–6

    2022

  28. [28]

    Semantic communication-assisted physical layer security over fading wiretap channels,

    X. Mu and Y . Liu, “Semantic communication-assisted physical layer security over fading wiretap channels,” inProc. IEEE Int. Conf. Commun. (ICC), Denver, CO, USA, Jun. 2024, pp. 2101–2106

    2024

  29. [29]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, and Z. Zhang, “Swin transformer: Hierarchical vision transformer using shifted windows,” inIEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Montreal, QC, Canada, Oct. 2021, pp. 9992–10002

    2021

  30. [30]

    Massive MIMO detection techniques: A survey,

    M. A. Albreem, M. Juntti, and S. Shahabuddin, “Massive MIMO detection techniques: A survey,”IEEE Commun. Surveys Tuts., vol. 21, no. 4, pp. 3109–3132, Aug. 2019

    2019

  31. [31]

    CBAM: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “CBAM: Convolutional block attention module,” inProc. Eur . Conf. on Comput. Vis. (ECCV), Munich, Germany, Sept. 2018, pp. 3–19

    2018

  32. [32]

    WITT: A wireless image transmission transformer for semantic communications,

    K. Yang, S. Wang, J. Dai, K. Tan, K. Niu, and P. Zhang, “WITT: A wireless image transmission transformer for semantic communications,” in IEEE Int. Conf. Acoust. Speech Signal Process. (ICASSP), Rhodes Island, Greece, Jun. 2023, pp. 1–5

    2023

  33. [33]

    Conditional variational image deraining,

    Y . Du, J. Xu, X. Zhen, M. -M. Cheng, and L. Shao, “Conditional variational image deraining,”IEEE Trans. Image Process., vol. 29, pp. 6288–6301, May 2020

    2020

  34. [34]

    Robust image semantic coding with learnable CSI fusion masking over MIMO fading channels,

    B. Xie, Y . Wu, Y . Shi, W. Zhang, S. Cui, and M. Debbah, “Robust image semantic coding with learnable CSI fusion masking over MIMO fading channels,”IEEE Trans. Wireless Commun., vol. 23, no. 10, pp. 14155– 14170, Oct. 2024

    2024

  35. [35]

    Learning structured output representation using deep conditional generative models,

    K. Sohn, X. Yan, and H. Lee, “Learning structured output representation using deep conditional generative models,”Adv. Neural Inf. Process. Syst. (NIPS), vol. 28, pp. 3483–3491, Dec. 2015

    2015

  36. [36]

    Variational inference: A review for statisticians,

    D. M. Blei, A. Kucukelbir, and J. D. McAuliffe, “Variational inference: A review for statisticians,”J. Amer . Stat. Assoc., vol. 112, no. 518, pp. 859–877, Jul. 2017

    2017

  37. [37]

    On information and sufficiency,

    S. Kullback and R. A. Leibler, “On information and sufficiency,”Ann. Math. Stat., pp. 79–86, 1951

    1951

  38. [38]

    Auto-Encoding Variational Bayes

    D. P. Kingma and M. Welling, “Auto-encoding variational bayes,” Dec. 2013, arXiv:1312.6114. [Online]. Available: http://arxiv.org/abs/1312.6114

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  39. [39]

    Stochastic successive convex approximation for non-convex constrained stochastic optimization,

    A. Liu, V . K. N. Lau, and B. Kananian, “Stochastic successive convex approximation for non-convex constrained stochastic optimization,”IEEE Trans. Signal Process., vol. 67, no. 16, pp. 4189–4203, Aug. 2019

    2019

  40. [40]

    The cityscapes dataset for semantic urban scene understanding,

    M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benenson, U. Franke, S. Roth, and B. Schiele, “The cityscapes dataset for semantic urban scene understanding,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV , USA, Jun. 2016, pp. 3213–3223

    2016

  41. [41]

    Semantic segmentation of outdoor panoramic images,

    S. Orhan and Y . Bastanlar, “Semantic segmentation of outdoor panoramic images,”Signal Image Video Process., vol. 16, pp. 643–650, Aug. 2021

    2021

  42. [42]

    Adam: A method for stochastic optimiza- tion,

    D. P. Kingma and J. L. Ba, “Adam: A method for stochastic optimiza- tion,” inProc. Int. Conf. Learn. Rep. (ICLR), San Diego, CA, USA, 2015, pp. 1–15

    2015

  43. [43]

    OFDM-guided deep joint source channel coding for wireless multipath fading channels,

    M. Yang, C. Bian, and H. -S. Kim, “OFDM-guided deep joint source channel coding for wireless multipath fading channels,”IEEE Trans. Cogn. Commun. Netw., vol. 8, no. 2, pp. 584–599, Jun. 2022

    2022