Abstract:Differentiable Neural Architecture Search (NAS) provides a promising avenue for automating the complex design of deep learning (DL) models. However, current differentiable NAS methods often face constraints in efficiency, operation selection, and adaptability under varying resource limitations. We introduce ZO-DARTS++, a novel NAS method that effectively balances performance and resource constraints. By integrating a zeroth-order approximation for efficient gradient handling, employing a sparsemax function with temperature annealing for clearer and more interpretable architecture distributions, and adopting a size-variable search scheme for generating compact yet accurate architectures, ZO-DARTS++ establishes a new balance between model complexity and performance. In extensive tests on medical imaging datasets, ZO-DARTS++ improves the average accuracy by up to 1.8\% over standard DARTS-based methods and shortens search time by approximately 38.6\%. Additionally, its resource-constrained variants can reduce the number of parameters by more than 35\% while maintaining competitive accuracy levels. Thus, ZO-DARTS++ offers a versatile and efficient framework for generating high-quality, resource-aware DL models suitable for real-world medical applications.
Abstract:Large Language Model (LLM) services often face challenges in achieving low inference latency and meeting Service Level Objectives (SLOs) under dynamic request patterns. Speculative decoding, which exploits lightweight models for drafting and LLMs for verification, has emerged as a compelling technique to accelerate LLM inference. However, existing speculative decoding solutions often fail to adapt to varying workloads and system environments, resulting in performance variability and SLO violations. In this paper, we introduce SpecServe, an efficient LLM inference system that dynamically adjusts speculative strategies according to real-time request loads and system configurations. SpecServe proposes a theoretical model to understand and predict the efficiency of speculative decoding across diverse scenarios. Additionally, it implements intelligent drafting and verification algorithms to guarantee optimal performance while achieving high SLO attainment. Experimental results on real-world LLM traces demonstrate that SpecServe consistently meets SLOs and achieves substantial performance improvements, yielding 1.14$\times$-14.3$\times$ speedups over state-of-the-art speculative inference systems.
Abstract:A common characteristic in integer linear programs (ILPs) is symmetry, allowing variables to be permuted without altering the underlying problem structure. Recently, GNNs have emerged as a promising approach for solving ILPs. However, a significant challenge arises when applying GNNs to ILPs with symmetry: classic GNN architectures struggle to differentiate between symmetric variables, which limits their predictive accuracy. In this work, we investigate the properties of permutation equivariance and invariance in GNNs, particularly in relation to the inherent symmetry of ILP formulations. We reveal that the interaction between these two factors contributes to the difficulty of distinguishing between symmetric variables. To address this challenge, we explore the potential of feature augmentation and propose several guiding principles for constructing augmented features. Building on these principles, we develop an orbit-based augmentation scheme that first groups symmetric variables and then samples augmented features for each group from a discrete uniform distribution. Empirical results demonstrate that our proposed approach significantly enhances both training efficiency and predictive performance.
Abstract:The impressive performance of ChatGPT and other foundation-model-based products in human language understanding has prompted both academia and industry to explore how these models can be tailored for specific industries and application scenarios. This process, known as the customization of domain-specific foundation models, addresses the limitations of general-purpose models, which may not fully capture the unique patterns and requirements of domain-specific data. Despite its importance, there is a notable lack of comprehensive overview papers on building domain-specific foundation models, while numerous resources exist for general-purpose models. To bridge this gap, this article provides a timely and thorough overview of the methodology for customizing domain-specific foundation models. It introduces basic concepts, outlines the general architecture, and surveys key methods for constructing domain-specific models. Furthermore, the article discusses various domains that can benefit from these specialized models and highlights the challenges ahead. Through this overview, we aim to offer valuable guidance and reference for researchers and practitioners from diverse fields to develop their own customized foundation models.
Abstract:Coordinated multipoint (CoMP) systems incur substantial resource consumption due to the management of backhaul links and the coordination among various base stations (BSs). Accurate prediction of coordinated spectral efficiency (CSE) can guide the optimization of network parameters, resulting in enhanced resource utilization efficiency. However, characterizing the CSE is intractable due to the inherent complexity of the CoMP channel model and the diversity of the 5G dynamic network environment, which poses a great challenge for CSE prediction in real-world 5G CoMP systems. To address this challenge, in this letter, we propose a data-driven model-assisted approach. Initially, we leverage domain knowledge to preprocess the collected raw data, thereby creating a well-informed dataset. Within this dataset, we explicitly define the target variable and the input feature space relevant to channel statistics for CSE prediction. Subsequently, a residual-based network model is built to capture the high-dimensional non-linear mapping function from the channel statistics to the CSE. The effectiveness of the proposed approach is validated by experimental results on real-world data.
Abstract:In this correspondence, we study deploying movable antenna (MA) array in a wideband multiple-input-single-output (MISO) communication system, where near-field (NF) channel model is considered. To alleviate beam squint effect, we propose to maximize the minimum analog beamforming gain across the entire wideband spectrum by appropriately adjusting MAs' positions, which is a highly challenging task. By introducing a slack variable and adopting the cutting-the-edge smoothed-gradient-descent-ascent (SGDA) method, we develop algorithms to resolve the aforementioned challenge. Numerical results verify the effectiveness of our proposed algorithms and demonstrate the benefit of utilizing MA array to mitigate beam squint effect in NF wideband system.
Abstract:The introduction of fifth-generation (5G) radio technology has revolutionized communications, bringing unprecedented automation, capacity, connectivity, and ultra-fast, reliable communications. However, this technological leap comes with a substantial increase in energy consumption, presenting a significant challenge. To improve the energy efficiency of 5G networks, it is imperative to develop sophisticated models that accurately reflect the influence of base station (BS) attributes and operational conditions on energy usage.Importantly, addressing the complexity and interdependencies of these diverse features is particularly challenging, both in terms of data processing and model architecture design. This paper proposes a novel 5G base stations energy consumption modelling method by learning from a real-world dataset used in the ITU 5G Base Station Energy Consumption Modelling Challenge in which our model ranked second. Unlike existing methods that omit the Base Station Identifier (BSID) information and thus fail to capture the unique energy fingerprint in different base stations, we incorporate the BSID into the input features and encoding it with an embedding layer for precise representation. Additionally, we introduce a novel masked training method alongside an attention mechanism to further boost the model's generalization capabilities and accuracy. After evaluation, our method demonstrates significant improvements over existing models, reducing Mean Absolute Percentage Error (MAPE) from 12.75% to 4.98%, leading to a performance gain of more than 60%.
Abstract:In this paper, we consider a cooperative sensing framework in the context of future multi-functional network with both communication and sensing ability, where one base station (BS) serves as a sensing transmitter and several nearby BSs serve as sensing receivers. Each receiver receives the sensing signal reflected by the target and communicates with the fusion center (FC) through a wireless multiple access channel (MAC) for cooperative target localization. To improve the localization performance, we present a hybrid information-signal domain cooperative sensing (HISDCS) design, where each sensing receiver transmits both the estimated time delay/effective reflecting coefficient and the received sensing signal sampled around the estimated time delay to the FC. Then, we propose to minimize the number of channel uses by utilizing an efficient Karhunen-Lo\'eve transformation (KLT) encoding scheme for signal quantization and proper node selection, under the Cram\'er-Rao lower bound (CRLB) constraint and the capacity limits of MAC. A novel matrix-inequality constrained successive convex approximation (MCSCA) algorithm is proposed to optimize the wireless backhaul resource allocation, together with a greedy strategy for node selection. Despite the high non-convexness of the considered problem, we prove that the proposed MCSCA algorithm is able to converge to the set of Karush-Kuhn-Tucker (KKT) solutions of a relaxed problem obtained by relaxing the discrete variables. Besides, a low-complexity quantization bit reallocation algorithm is designed, which does not perform explicit node selection, and is able to harvest most of the performance gain brought by HISDCS. Finally, numerical simulations are presented to show that the proposed HISDCS design is able to significantly outperform the baseline schemes.
Abstract:Accurate classification of medical images is essential for modern diagnostics. Deep learning advancements led clinicians to increasingly use sophisticated models to make faster and more accurate decisions, sometimes replacing human judgment. However, model development is costly and repetitive. Neural Architecture Search (NAS) provides solutions by automating the design of deep learning architectures. This paper presents ZO-DARTS+, a differentiable NAS algorithm that improves search efficiency through a novel method of generating sparse probabilities by bi-level optimization. Experiments on five public medical datasets show that ZO-DARTS+ matches the accuracy of state-of-the-art solutions while reducing search times by up to three times.
Abstract:Semantic communication (SemCom) is emerging as a key technology for future sixth-generation (6G) systems. Unlike traditional bit-level communication (BitCom), SemCom directly optimizes performance at the semantic level, leading to superior communication efficiency. Nevertheless, the task-oriented nature of SemCom renders it challenging to completely replace BitCom. Consequently, it is desired to consider a semantic-bit coexisting communication system, where a base station (BS) serves SemCom users (sem-users) and BitCom users (bit-users) simultaneously. Such a system faces severe and heterogeneous inter-user interference. In this context, this paper provides a new semantic-bit coexisting communication framework and proposes a spatial beamforming scheme to accommodate both types of users. Specifically, we consider maximizing the semantic rate for semantic users while ensuring the quality-of-service (QoS) requirements for bit-users. Due to the intractability of obtaining the exact closed-form expression of the semantic rate, a data driven method is first applied to attain an approximated expression via data fitting. With the resulting complex transcendental function, majorization minimization (MM) is adopted to convert the original formulated problem into a multiple-ratio problem, which allows fractional programming (FP) to be used to further transform the problem into an inhomogeneous quadratically constrained quadratic programs (QCQP) problem. Solving the problem leads to a semi-closed form solution with undetermined Lagrangian factors that can be updated by a fixed point algorithm. Extensive simulation results demonstrate that the proposed beamforming scheme significantly outperforms conventional beamforming algorithms such as zero-forcing (ZF), maximum ratio transmission (MRT), and weighted minimum mean-square error (WMMSE).