Institution profile

Shenzhen Research Institute of Big Data

Academic institutionasia · cn
Official website
Research library147linked papers
Opportunities0open roles
Selected work

Representative Papers

Suppressing Beam Squint Effect For Near-Field Wideband Communication Through Movable Antennas

Jul 28, 2024arXiv.org

In near-field broadband MISO systems, movable-antenna (MA) arrays suffer from beam squint, causing inconsistent analog beamforming gain across the entire bandwidth. To address this, this work pioneers the use of physical antenna displacement as an additional controllable degree of freedom. We propose a joint optimization framework grounded in near-field channel modeling, aiming to maximize the minimum analog beamforming gain over the full bandwidth. The problem is solved via a synergistic combination of the smoothed gradient descent-ascent (SGDA) algorithm and a relaxation-variable technique. Unlike conventional fixed-array designs, our approach dynamically reconfigures antenna positions to mitigate squint-induced gain variation. Under typical near-field conditions, it achieves up to a 3.2-dB improvement in the minimum broadband beam gain, significantly enhancing gain consistency and robustness. This establishes a novel paradigm for broadband near-field beamforming.

3 citationsRead paper

Constant Bit-size Transformers Are Turing Complete

May 22, 2025arXiv.org

This work addresses the fundamental question of whether constant-bit-width Transformers can simulate arbitrary Turing machines and precisely characterizes their computational expressiveness. We introduce a novel simulation framework based on Post machines (queue automata), leveraging formal queue-behavior abstraction and rigorous computability-theoretic analysis. We establish, for the first time, that a constant-bit Transformer with context window length $ s(n) $ can simulate all computations in $ ext{SPACE}[s(n)] $, and achieves Turing completeness when $ s(n) = Omega(log n) $. Crucially, this result breaks the conventional assumption that expressive power necessitates scaling model parameters or numerical precision with input size. Our analysis yields the first exact equivalence between constant-bit Transformers and a classical space-complexity class, thereby providing a new theoretical foundation for understanding the intrinsic reasoning capabilities of Transformer architectures.

1 citationsRead paper

A Hybrid Autoencoder-Transformer Model for Robust Day-Ahead Electricity Price Forecasting under Extreme Conditions

Apr 25, 20252025 IEEE International Symposium on the Application of Artificial Intelligence in Electrical Engineering (AAIEE)

To address the low accuracy and poor robustness of day-ahead electricity price forecasting (DAEPF) under extreme weather and market anomalies, this paper proposes a Distillation-Attention Transformer with Autoencoder-based Self-Regression (DAT-ASR) framework. The method innovatively integrates a dynamic attention weight allocation mechanism and an unsupervised anomaly pattern separation module to jointly capture long- and short-term price dependencies while explicitly modeling anomalous disturbances. Furthermore, self-supervised pretraining enhances generalization in low-data anomaly scenarios. Experiments on California ISO and Shandong Power Market datasets demonstrate that DAT-ASR achieves 12.7%–18.3% lower average MAE than state-of-the-art models, reduces prediction errors during anomalous periods by over 25%, and improves inference speed by approximately 40%. These results substantiate significant gains in DAEPF accuracy, robustness, and practicality under extreme conditions.

1 citationsRead paper
Recent publications

Latest Papers

Error-Aware Reverse Auction Mechanism for Large Language Model Routing

Aug 12, 2026

This work addresses the limitations of centralized prediction-based routing in large language models (LLMs), which often leads to misaligned information risks and scalability bottlenecks. To overcome these issues, the paper introduces, for the first time, a reverse auction mechanism into LLM routing and proposes the error-aware EA-RAM framework. In this framework, model providers autonomously bid their success rates and costs, while explicit modeling of dual sources of noise—arising from both prediction and evaluation—enables robust handling of uncertainty. The mechanism is shown to be Bayesian incentive-compatible and individually rational, with a provable upper bound on social welfare loss. Empirical results demonstrate that EA-RAM consistently outperforms centralized baselines across both simulated and real-world benchmarks, maintaining robustness under dual-error conditions and significantly advancing the cost-performance Pareto frontier.

0 citationsRead paper

Active Perception for Embodied Disambiguation

Aug 11, 2026

This study addresses target ambiguity in embodied environments caused by insufficient physical evidence by proposing an active perception framework grounded in vision-language models. Treating active observation as the primary mechanism for information acquisition, the method unifies physical exploration and user interaction within a single disambiguation process, enabling robots to autonomously decide between executing active observations or querying users to clarify intent. Real-world experiments demonstrate that this framework effectively integrates environmental exploration with human-robot interaction, significantly enhancing both target disambiguation performance and task success rates in complex scenarios.

0 citationsRead paper