arXiv subjects

Haochuan Zhang

Publications and source records attributed to Haochuan Zhang.

11 recordsLinked to original sources

arXiv2025

UMind: A Unified Multitask Network for Zero-Shot M/EEG Visual Decoding

Decoding visual information from time-resolved brain recordings, such as EEG and MEG, plays a pivotal role in real-time brain-computer interfaces. However, existing approaches primarily focus on direct brain-image feature alignment and are limited to single-task frameworks or task-specific models. In this paper, we propose a Unified MultItask Network for zero-shot M/EEG visual Decoding (referred to UMind), including visual stimulus retrieval, classification, and reconstruction, where multiple tasks mutually enhance each other. Our method learns robust neural-visual and semantic representations through multimodal alignment with both image and text modalities. The integration of both coarse and fine-grained texts enhances the extraction of these neural representations, enabling more detailed semantic and visual decoding. These representations then serve as dual conditional inputs to a pre-trained diffusion model, guiding visual reconstruction from both visual and semantic perspectives. Extensive evaluations on MEG and EEG datasets demonstrate the effectiveness, robustness, and biological plausibility of our approach in capturing spatiotemporal neural dynamics. Our approach sets a multitask pipeline for brain visual decoding, highlighting the synergy of semantic information in visual feature extraction. The code is available at https://github.com/xuchengjian632/UMind.

cs.HC↗

arXiv2025

TempoGPT: Enhancing Time Series Reasoning via Quantizing Embedding

Multi-modal language model has made advanced progress in vision and audio, but still faces significant challenges in dealing with complex reasoning tasks in the time series domain. The reasons are twofold. First, labels for multi-modal time series data are coarse and devoid of analysis or reasoning processes. Training with these data cannot improve the model's reasoning capabilities. Second, due to the lack of precise tokenization in processing time series, the representation patterns for temporal and textual information are inconsistent, which hampers the effectiveness of multi-modal alignment. To address these challenges, we propose a multi-modal time series data construction approach and a multi-modal time series language model (TLM), TempoGPT. Specially, we construct multi-modal data for complex reasoning tasks by analyzing the variable-system relationships within a white-box system. Additionally, proposed TempoGPT achieves consistent representation between temporal and textual information by quantizing temporal embeddings, where temporal embeddings are quantized into a series of discrete tokens using a predefined codebook; subsequently, a shared embedding layer processes both temporal and textual tokens. Extensive experiments demonstrate that TempoGPT accurately perceives temporal information, logically infers conclusions, and achieves state-of-the-art in the constructed complex time series reasoning tasks. Moreover, we quantitatively demonstrate the effectiveness of quantizing temporal embeddings in enhancing multi-modal alignment and the reasoning capabilities of TLMs. Code and data are available at https://github.com/zhanghaochuan20/TempoGPT.

cs.LG↗

arXiv2025

Sensorformer: Cross-patch attention with global-patch compression is effective for high-dimensional multivariate time series forecasting

Among the existing Transformer-based multivariate time series forecasting methods, iTransformer, which treats each variable sequence as a token and only explicitly extracts cross-variable dependencies, and PatchTST, which adopts a channel-independent strategy and only explicitly extracts cross-time dependencies, both significantly outperform most Channel-Dependent Transformer that simultaneously extract cross-time and cross-variable dependencies. This indicates that existing Transformer-based multivariate time series forecasting methods still struggle to effectively fuse these two types of information. We attribute this issue to the dynamic time lags in the causal relationships between different variables. Therefore, we propose a new multivariate time series forecasting Transformer, Sensorformer, which first compresses the global patch information and then simultaneously extracts cross-variable and cross-time dependencies from the compressed representations. Sensorformer can effectively capture the correct inter-variable correlations and causal relationships, even in the presence of dynamic causal lags between variables, while also reducing the computational complexity of pure cross-patch self-attention from $O(D^2 \cdot Patch\_num^2 \cdot d\_model)$ to $O(D^2 \cdot Patch\_num \cdot d\_model)$. Extensive comparative and ablation experiments on 9 mainstream real-world multivariate time series forecasting datasets demonstrate the superiority of Sensorformer. The implementation of Sensorformer, following the style of the Time-series-library and scripts for reproducing the main results, is publicly available at https://github.com/BigYellowTiger/Sensorformer

cs.LG↗

arXiv2024

Vector Approximate Survey Propagation

Approximate Message Passing (AMP), originally designed to solve high-dimensional linear inverse problems, has found broad applications in signal processing and statistical inference. Among its key variants, Vector Approximate Message Passing (VAMP) and Generalized Approximate Survey Propagation (GASP) have demonstrated effectiveness even in scenarios where the assumed generative models differ from the true models. However, the maximum a posteriori (MAP) versions of VAMP and GASP have limitations: VAMP is restricted to differentiable priors and likelihoods, while GASP requires the measurement matrix to have independent identically distributed (i.i.d.) elements. To overcome these limitations, this paper introduces a new algorithm, Vector Approximate Survey Propagation (VASP). VASP utilizes survey propagation to handle non-differentiable priors and likelihoods, along with employs vector-form messages to account for correlations in the measurement matrix. Simulations reveal that VASP significantly surpasses VAMP and GASP in estimation accuracy, particularly when the assumed prior is discrete-supported and the measurement matrix is non-i.i.d.. Additionally, the state evolution (SE) of VASP, derived heuristically, accurately reflects the per-iteration mean squared error (MSE). A comparison between the SE and the free energy computed by Takahashi and Kabashima under the one-step replica symmetry breaking (1RSB) ansatz shows that the SE's fixed-point equations align with the free energy's saddle point equations. This suggests that VASP efficiently implements the postulated MAP estimator (which is NP-hard in the worst case) with cubic computational complexity, assuming the 1RSB ansatz is valid.

cs.IT↗

arXiv2024

K-step Vector Approximate Survey Propagation

Approximate Message Passing (AMP), originally developed to address high-dimensional linear inverse problems, has found widespread applications in signal processing and statistical inference. Among its notable variants, Vector Approximate Message Passing (VAMP), Generalized Approximate Survey Propagation (GASP), and Vector Approximate Survey Propagation (VASP) have demonstrated effectiveness even when the assumed generative models differ from the true models. However, many fundamental questions regarding model mismatch remain unanswered. For instance, it is still unclear what level of model mismatch is required for the postulated posterior estimate (PPE) to exhibit a replica symmetry breaking (RSB) structure in the extremum conditions of its free energy, and what order of RSB is necessary. In this paper, we introduce a novel approximate message passing algorithm that incorporates K-step RSB (KRSB) and naturally reduces to VAMP and VASP with specific parameter selections. We refer to this as the K-step VASP (KVASP) algorithm. Simulations show that KVASP significantly outperforms VAMP and GASP in estimation accuracy, particularly when the assumed prior has discrete support and the measurement matrix is non-i.i.d.. Additionally, the state evolution (SE) of KVASP, derived heuristically, accurately tracks the per-iteration mean squared error (MSE). A comparison between the SE and the free energy under the KRSB ansatz reveals that the fixed-point equations of SE align with the saddle-point equations of the free energy. This suggests that, once the KRSB ansatz holds and the SE fixed point is reached, KVASP can accurately compute the PPE in the large system limit (LSL).

cs.IT↗

arXiv2024

Symbol Detection for Coarsely Quantized OTFS

This paper explicitly models a coarse and noisy quantization in a communication system empowered by orthogonal time frequency space (OTFS) for cost and power efficiency. We first point out, with coarse quantization, the effective channel is imbalanced and thus no longer able to circularly shift the transmitted symbols along the delay-Doppler domain. Meanwhile, the effective channel is non-isotropic, which imposes a significant loss to symbol detection algorithms like the original approximate message passing (AMP). Although the algorithm of generalized expectation consistent for signal recovery (GEC-SR) can mitigate this loss, the complexity in computation is prohibitively high, mainly due to an dramatic increase in the matrix size of OTFS. In this context, we propose a low-complexity algorithm that incorporates into the GEC-SR a quick inversion of quasi-banded matrices, reducing the complexity from a cubic order to a linear order while keeping the performance at the same level.

cs.IT↗

arXiv2021

Multi-Layer Bilinear Generalized Approximate Message Passing

In this paper, we extend the bilinear generalized approximate message passing (BiG-AMP) approach, originally proposed for high-dimensional generalized bilinear regression, to the multi-layer case for the handling of cascaded problem such as matrix-factorization problem arising in relay communication among others. Assuming statistically independent matrix entries with known priors, the new algorithm called ML-BiGAMP could approximate the general sum-product loopy belief propagation (LBP) in the high-dimensional limit enjoying a substantial reduction in computational complexity. We demonstrate that, in large system limit, the asymptotic MSE performance of ML-BiGAMP could be fully characterized via a set of simple one-dimensional equations termed state evolution (SE). We establish that the asymptotic MSE predicted by ML-BiGAMP' SE matches perfectly the exact MMSE predicted by the replica method, which is well known to be Bayes-optimal but infeasible in practice. This consistency indicates that the ML-BiGAMP may still retain the same Bayes-optimal performance as the MMSE estimator in high-dimensional applications, although ML-BiGAMP's computational burden is far lower. As an illustrative example of the general ML-BiGAMP, we provide a detector design that could estimate the channel fading and the data symbols jointly with high precision for the two-hop amplify-and-forward relay communication systems.

cs.IT↗

arXiv2021

Expectation-Maximization-Aided Hybrid Generalized Expectation Consistent for Sparse Signal Reconstruction

The reconstruction of sparse signal is an active area of research. Different from a typical i.i.d. assumption, this paper considers a non-independent prior of group structure. For this more practical setup, we propose EM-aided HyGEC, a new algorithm to address the stability issue and the hyper-parameter issue facing the other algorithms. The instability problem results from the ill condition of the transform matrix, while the unavailability of the hyper-parameters is a ground truth that their values are not known beforehand. The proposed algorithm is built on the paradigm of HyGAMP (proposed by Rangan et al.) but we replace its inner engine, the GAMP, by a matrix-insensitive alternative, the GEC, so that the first issue is solved. For the second issue, we take expectation-maximization as an outer loop, and together with the inner engine HyGEC, we learn the value of the hyper-parameters. Effectiveness of the proposed algorithm is also verified by means of numerical simulations.

cs.IT↗

arXiv2020

Estimation for High-Dimensional Multi-Layer Generalized Linear Model -- Part I: The Exact MMSE Estimator

This two-part work considers the minimum means square error (MMSE) estimation problem for a high dimensional multi-layer generalized linear model (ML-GLM), which resembles a feed-forward fully connected deep learning network in that each of its layer mixes up the random input with a known weighting matrix and activates the results via non-linear functions, except that the activation here is stochastic and following some random distribution. Part I of the work focuses on the exact MMSE estimator, whose implementation is long known infeasible. For this exact estimator, an asymptotic analysis on the performance is carried out using a new replica method that is refined from certain aspects. A decoupling principle is then established, suggesting that, in terms of joint input-and-estimate distribution, the original estimation problem of multiple-input multiple-output is indeed identical to a simple single-input single-output one subjected to additive white Gaussian noise (AWGN) only. The variance of the AWGN is further shown to be determined by some coupled equations, whose dependency on the weighting and activation is given explicitly and analytically. Comparing to existing results, this paper is the first to offer a decoupling principle for the ML-GLM estimation problem. To further address the implementation issue of an exact solution, Part II proposes an approximate estimator, ML-GAMP, whose per-iteration complexity is as low as GAMP, while its asymptotic MSE (if converged) is as optimal as the exact MMSE estimator.

cs.IT↗

arXiv2020

Estimation for High-Dimensional Multi-Layer Generalized Linear Model -- Part II: The ML-GAMP Estimator

This is Part II of a two-part work on the estimation for a multi-layer generalized linear model (ML-GLM) in large system limits. In Part I, we had analyzed the asymptotic performance of an exact MMSE estimator, and obtained a set of coupled equations that could characterize its MSE performance. To work around the implementation difficulty of the exact estimator, this paper continues to propose an approximate solution, ML-GAMP, which could be derived by blending a moment-matching projection into the Gaussian approximated loopy belief propagation. The ML-GAMP estimator is then shown to enjoy a great simplicity in its implementation, where its per-iteration complexity is as low as GAMP. Further analysis on its asymptotic performance also reveals that, in large system limits, its dynamical MSE behavior is fully characterized by a set of simple one-dimensional iterating equations, termed state evolution (SE). Interestingly, this SE of ML-GAMP share exactly the same fixed points with an exact MMSE estimator whose fixed points were obtained in Part I via a replica analysis. Given the Bayes-optimality of the exact implementation, this proposed estimator (if converged) is optimal in the MSE sense.

cs.IT↗

arXiv2020

Symbol Detection for Massive MIMO AF Relays Using Approximate Bayesian Inference

For massive MIMO AF relays, symbol detection becomes a practical issue when the number of antennas is not large enough, since linear methods are non-optimal and optimal methods are exponentially complex. This paper proposes a new detection algorithm that offers Bayesian-optimal MSE at the cost of $O(n^3)$ complexity per iteration. The algorithm is in essence a hybrid of two methods recently developed for deep learning, with particular optimization for relay. As a hybrid, it inherits from the two a state evolution formulism, where the asymptotic MSE can be precisely predicted through a scalar equivalent model. The algorithm also degenerates easily to many results well-known when single-hop considered.

cs.IT↗