Xinjue Wang

xjw [at] ieee.org

Google Scholar GitHub

Hi!

I obtained my Ph.D. in Electrical Engineering in 2026 at Aalto University, advised by Professors Esa Ollila and Sergiy A. Vorobyov. I received my M.Sc. in Electrical Engineering from Aalto University in 2022, and my B.Eng. in Information Engineering from Northwestern Polytechnical University, Xi'an, China, in 2020.

My work sits between statistical signal processing and optimization, and runs along two lines.

1. Robustness and reliability. Both parts below ask the same question: what happens to a method when the assumptions behind it do not hold.

1.1 Under heavy-tailed noise. Estimators built on a Gaussian noise model degrade sharply when the noise is impulsive or heavy-tailed. The classical robust alternatives avoid that failure but lose efficiency in the nominal case. I design M-estimation and covariance-matching procedures that keep both, with convergence guarantees for the iterations they run on. Recent work covers device activity detection in massive random access, along with its mixed near/far-field and algorithm-unfolded variants.

1.2 Trustworthy AI. A graph neural network is trained on a graph that is rarely exact: edges are noisy, missing, or change over time. I model those perturbations probabilistically and use spectral analysis to bound how far the network's output can move, and to characterize how the error propagates layer by layer. The result is a way to estimate the effect of a given level of graph noise before the model is deployed.

2. Low-rank structure. In high dimensions the data alone is not enough to identify the model, and structural assumptions have to make up the difference. I use low-rank and tensor factorizations in two ways. As priors they make ill-posed inverse problems identifiable, which is the basis of my work on tensor regression and hyperspectral deconvolution. As budget constraints they reduce the cost of adapting large models, which I study with CP tensor adapters and fixed-rank manifold methods for fine-tuning.

Publications

Preprints & Under Review

  1. Diffusion-Based Distributionally Robust Adaptive Beamforming

    K.H. Irani, X. Wang, and S.A. Vorobyov. Sept. 2026.

  2. Robust Parallel Covariance Learning for Activity Detection in Massive Random Access

    X. Wang, S.A. Vorobyov, and E. Ollila. Sept. 2026.

  3. Parallel Robust Covariance Learning with Algorithm Unfolding for Activity Detection

    X. Wang, S.A. Vorobyov, and E. Ollila. Aug. 2026.

  4. Zeroth-Order Riemannian Optimization on Fixed-Rank Manifolds for LLM Fine-Tuning

    X. Wang, X. Wang, Y. Zhang, S.A. Vorobyov, and E. Ollila. May 2026.

Journal Papers

  1. A Majorization-Minimization Framework for Activity Detection in Mixed Near-and-Far-Field Random Access

    X. Wang, Z.-Y. Wang, S.A. Vorobyov, E. Ollila, G.A.A. Baduge, and M. Vaezi. IEEE Transactions on Wireless Communications, vol. 25, pp. – , early access, 2026.

  2. Generalized Nonnegative Structured Kruskal Tensor Regression (arXiv) (code)

    X. Wang, E. Ollila, S.A. Vorobyov, and A. Mian. Signal Processing, vol. 240, 110338, Mar. 2026.

  3. Robust Activity Detection for Massive Random Access (arXiv) (code)

    X. Wang, E. Ollila, and S.A. Vorobyov. IEEE Transactions on Signal Processing, vol. 73, pp. 3513–3527, Aug. 2025.

  4. Graph Convolutional Neural Networks Sensitivity under Probabilistic Error Model (arXiv)

    X. Wang, E. Ollila, and S.A. Vorobyov. IEEE Transactions on Signal and Information Processing over Networks, vol. 10, pp. 788–803, Oct. 2024.

  5. Tracking the Occluded Indoor Target with Scattered Millimeter Wave Signal

    Y. Xu, X. Wang, J. Kupiainen, J. Sae, J. Boutellier, J. Nurmi, and B. Tan. IEEE Sensors Journal, vol. 24, no. 22, pp. 38102–38112, Nov. 2024.

Conference Papers

  1. GeoMix: Descriptor-Free Visual Localization via Global Context and Multi-Detector Training (arXiv)

    Y. Zhang, X. Wang, Z. Wang, E. Rahtu, and J. Kannala. ECCV 2026, Malmö, Sweden, Sept. 8–12, 2026, pp. 1–20.

  2. Covariance-Aware MM-PGD for Mixed Near/Far-Field Activity Detection (arXiv)

    X. Wang, Z.-Y. Wang, S.A. Vorobyov, E. Ollila, G.A.A. Baduge, and M. Vaezi. SPAWC 2026, Athens, Greece, Sept. 7–9, 2026.

  3. Finer Parameter Steps for Low-Rank PEFT: A Controlled Study with CP Tensor Adapters (arXiv)

    X. Wang, X. Wang, Y. Zhang, S.A. Vorobyov, E. Ollila, and Z.-Y. Wang. ICML 2026 Workshop CoLoRAI, Seoul, South Korea, July 10, 2026, pp. 1–8.

  4. Anisotropic Tensor Deconvolution of Hyperspectral Images (arXiv) (code)

    X. Wang, X. Wang, E. Ollila, and S.A. Vorobyov. ICASSP 2026, Barcelona, Spain, May 2026, pp. 706–710.

  5. Robust Activity Detection for Massive Access Using Covariance-Based Matching Pursuit

    X. Wang, E. Ollila, and S.A. Vorobyov. ICASSP 2025, Hyderabad, India, Apr. 2025, pp. 1–5.

  6. Nonnegative Structured Kruskal Tensor Regression

    X. Wang, E. Ollila, and S.A. Vorobyov. CAMSAP 2023 (invited paper), Herradura, Costa Rica, Dec. 2023, pp. 441–445.

  7. Correlation-Based Graph Smoothness Measures in Graph Signal Processing

    J. Miettinen, S.A. Vorobyov, E. Ollila, and X. Wang. EUSIPCO 2023, Helsinki, Finland, Sept. 2023, pp. 1848–1852.

  8. Graph Neural Network Sensitivity under Probabilistic Error Model

    X. Wang, E. Ollila, and S.A. Vorobyov. EUSIPCO 2022, Belgrade, Serbia, Aug. 2022, pp. 2146–2150.

  9. Distributed Principal Component Analysis Based on Randomized Low-Rank Approximation

    X. Wang and J. Chen. ICSPCC 2020, Macau, China, Aug. 2020, pp. 1–5.

Academic Service

Journal referee: IEEE Trans. Inf. Theory, IEEE Trans. Wireless Commun., IEEE Trans. Mach. Learn. Commun. Netw., Scand. J. Stat.

Conference reviewer: ICASSP 2025, 2026, IJCNN 2025, ITW 2024

Education