The surge in large language models (LLMs) has fundamentally reshaped the landscape of GPU usage patterns, creating an urgent need for more efficient management strategies. While cloud providers employ spot instances to reduce costs for low-priority (LP) tasks, existing schedulers still grapple with high eviction rates and lengthy queuing times. To address these limitations, we present GFS, a novel preemptive scheduling framework that enhances service-level objective (SLO) compliance for high-priority (HP) tasks while minimizing preemptions to LP tasks. Firstly, GFS utilizes a lightweight forecasting model that predicts GPU demand among different tenants, enabling proactive resource management. Secondly, GFS employs a dynamic allocation mechanism to adjust the spot quota for LP tasks with guaranteed durations. Lastly, GFS incorporates a preemptive scheduling policy that prioritizes HP tasks while minimizing the impact on LP tasks. We demonstrate the effectiveness of GFS through both real-world implementation and simulations. The results show that GFS reduces eviction rates by 33.0%, and cuts queuing delays by 44.1% for LP tasks. Furthermore, GFS enhances the GPU allocation rate by up to 22.8% in real production clusters. In a production cluster of more than 10,000 GPUs, GFS yields roughly $459,715 in monthly benefits.

This paper introduces delay-alignment modulation (DAM) for secure integrated sensing and communication (ISAC). Due to the broadcast nature of multi-user downlinks, communications are vulnerable to eavesdropping. DAM applies controlled per-path symbol delays at the transmitter to coherently align the multipath components at the intended user, enhancing the received signal power, while simultaneously creating delay misalignment at the eavesdropper (Eve). To mitigate sensing degradation caused by multipath propagation, we propose a two-stage protocol that first estimates the angle and then the delay of the line-of-sight (LoS) path after suppressing multipath interference. We derive the secrecy spectral efficiency (SSE) and the Cramer–Rao bound (CRB) of the target delay. Finally, we develop a path-based zero-forcing (ZF) precoding framework and formulate a max–min SSE design under CRB and power constraints. Simulation results show DAM significantly outperforms the strongest-path (SP) benchmark in terms of SSE, while meeting sensing requirements, since intentional delay alignment at legitimate users degrades Eve’s reception.

This paper proposes a novel optimization framework for enhancing the security resilience of cell-free massive multiple-input multiple-output (CF-mMIMO) networks with multi-antenna access points (APs) and protective partial zero-forcing (PPZF) under active eavesdropping. Based on the main principles of absorption, adaptation, and recovery, we formulate a security aware resilience metric to quantify the system performance during and after a security outage. A multi-user service priority-aware power allocation problem is formulated to minimize the mean squared error (MSE) between real-time and desired security efficiency, thereby enabling a trade-off between the target user’s secrecy performance and multi-user quality of service (QoS). To solve this non-convex problem, a security-aware iterative algorithm based on the successive convex approximation (SCA) is employed. The proposed algorithm determines the optimal power allocation strategy by balancing solution quality against recovery time. At each iteration, it evaluates the overall resilience score and selects the strategy that achieves the highest value. Simulation results confirm that the proposed framework significantly improves the resilience of CF-mMIMO networks, allowing flexible adaptation between rapid recovery and high-quality recovery, depending on system requirements.

ABSTRACT Urban transportation systems are essential in fulfilling the requirements of residents while guaranteeing the normal functioning of cities. These systems encompass various modes of transportation, infrastructure, and services that enable people to move within and between urban areas. However, the challenges posed by escalating urbanization, particularly the growing menace of traffic congestion, underscore the pressing need for effective solutions. We propose an Adaptive and Dynamic Spatio‐Temporal Network (ADSTN) as an innovative solution to the complexities associated with traffic congestion. The identified shortcomings in existing models, such as their limitations in capturing authentic spatial dependencies, insufficient understanding of the heterogeneous relationship between the temporal and spatial domains, and the oversight of local trend information, motivate the development of ADSTN. The model integrates three key components: a learnable adaptive attention module, a local temporal self‐attention block, and a spatio‐temporal dynamic graph convolution block. ADSTN stands out for its outstanding performance in handling local spatio‐temporal dependencies, periodicity, and dynamics within traffic flow forecasting. Evaluation on three public real‐world datasets underscores the competitive achievements of ADSTN contrasted with state‐of‐the‐art models, all while maintaining computational efficiency.

With the substantial increase in the number of antennas, polar-domain channel modeling for extremely large-scale antenna array (ELAA) systems has been introduced to capture both angular and distance information in near-field environments. The fine-grained polar-domain channel provides additional sources of randomness, making it well-suited for physical layer key generation (PLKG). To minimize the pilot overhead in multi-user key generation and leverage the randomness from the polar-domain channel paths, we herein design a zero-forcing (ZF)-based precoding scheme to mitigate inter-path and inter-user interference. Using ZF precoding, we derive an analytical expression for the sum secret key rate (SKR) as a function of power allocation variables, and then optimize these variables in the presence of eavesdroppers. Our simulations validate the proposed precoding design and power allocation methods in terms of sum SKR versus the transmit power, antenna configurations, and spatial correlation between legitimate and eavesdropping channels.

Extremely large-scale antenna arrays (ELAA) require near-field spherical wave modeling due to the substantial increase in the number of antennas, which introduces new spatial dimensions to physical layer key generation (PLKG). We investigate multi-user PLKG in near-field environments, where a base station with an ELAA simultaneously generates secret keys with multiple users. We derive an analytical expression for the key rate (KR). By utilizing spatial dimensions of distance and angle in near-field environments, we apply eigenvalue decomposition and singular value decomposition to design precoding matrices to reduce interference among user equipments (UEs) and extract uncorrelated subchannels. Given that the KR is non-convex, we approximate it and optimize the precoding matrix to increase the KR. After precoding design, the KR depends on the transmit power allocated to the subchannels. Two optimization problems are formulated to further optimize transmit power allocation. The first problem focuses on maximizing the sum KR. We apply the Lagrange multiplier method to determine the optimal power allocation variables by searching the Lagrange multiplier. To reduce computational complexity, a supervised feedforward neural network (FNN) is designed to capture the relationship between the power allocation variables and the Lagrange multiplier. The second optimization problem focuses on KR fairness. By introducing a slack variable that is smaller than the KRs of all users, we use the CVX toolbox to find optimal power allocation variables that maximize this slack variable. To further reduce complexity, the Lagrange multiplier method offers an analytical solution for power allocation variables in terms of Lagrange multipliers determined by the slack variable in the high-power case. We employ a bisection algorithm to find the slack variable. Furthermore, we propose an FNN to map transmit power to the slack variable. Simulations demonstrate that the proposed methods efficiently leverage near-field effects for multi-user PLKG, reducing pilot overhead.

Protein language models have demonstrated significant potential in the field of protein engineering. However, current protein language models primarily operate at the residue scale, which limits their ability to provide information at the atom level. This limitation prevents us from fully exploiting the capabilities of protein language models for applications involving both proteins and small molecules. In this paper, we propose ESM-AA (ESM All-Atom), a novel approach that enables atom-scale and residue-scale unified molecular modeling. ESM-AA achieves this by pre-training on multi-scale code-switch protein sequences and utilizing a multi-scale position encoding to capture relationships among residues and atoms. Experimental results indicate that ESM-AA surpasses previous methods in protein-molecule tasks, demonstrating the full utilization of protein language models. Further investigations reveal that through unified molecular modeling, ESM-AA not only gains molecular knowledge but also retains its understanding of proteins.

Physical-layer key generation (PKG) based on wireless channels is a lightweight technique to establish secure keys between legitimate communication nodes. Recently, intelligent reflecting surfaces (IRSs) have been leveraged to enhance the performance of PKG in terms of secret key rate (SKR), as it can reconfigure the wireless propagation environment and introduce more channel randomness. In this paper, we investigate an IRS-assisted PKG system, taking into account the channel spatial correlation at both the base station (BS) and the IRS. Based on the considered system model, the closed-form expression of SKR is derived analytically considering correlated eavesdropping channels. Aiming to maximize the SKR, a joint design problem of the BS’s precoding matrix and the IRS’s phase shift vector is formulated. To address this high-dimensional non-convex optimization problem, we propose a novel unsupervised deep neural network (DNN)-based algorithm with a simple structure. Different from most previous works that adopt iterative optimization to solve the problem, the proposed DNN-based algorithm directly obtains the BS precoding and IRS phase shifts as the output of the DNN. Simulation results reveal that the proposed DNN-based algorithm outperforms the benchmark methods with regard to SKR.

Physical layer key generation (PLKG) leverages wireless channels to produce secret keys for legitimate users. However, in millimetre-wave (mmWave) frequency bands, the presence of blockage significantly reduces the key rate (KR) of a PLKG system. To address this issue, we introduce reconfigurable intelligent surfaces (RISs) as a potential solution for constructing RIS-reflected channels, thereby enhancing the KR. Our study focuses on the beam-domain channel model and exploits the sparsity of mmWave bands to enhance the randomness of secret keys. To relieve pilot overhead in multi-user systems, we employ a compressed sensing (CS) algorithm to estimate angular information and propose a channel probing protocol with the full-array configuration for acquiring the beam-domain channel. We derive the analytical expressions for the KR in the case of full-array configuration. To optimize the KR, we design the phase shift and precoding vectors based on the obtained angular information. Furthermore, we employ a water-filling algorithm that relies on the Karush-Kuhn-Tucker (KKT) conditions to optimize power allocation for estimating the beam-domain channel with the same channel variance. When channel variances of the beam-domain channel differ, we design a deep-learning-based power allocation method for a more complex problem. What is more, we design a sub-array configuration scheme that exploits the difference in spatial angles between users to reduce pilot overhead and derive the analytical expression for the KR. Through extensive simulations, we demonstrate that our proposed PLKG schemes outperform existing methods.

Physical-layer key generation (PKG) based on wireless channels is a lightweight technique to establish secure keys between legitimate communication nodes. Recently, intelligent reflecting surfaces (IRSs) have been leveraged to enhance the performance of PKG in terms of secret key rate (SKR), as it can reconfigure the wireless propagation environment and introduce more channel randomness. In this paper, we investigate an IRS-assisted PKG system, taking into account the channel spatial correlation at both the base station (BS) and the IRS. Based on the considered system model, the closed-form expression of SKR is derived analytically. Aiming to maximize the SKR, a joint design problem of the BS's precoding matrix and the IRS's reflecting coefficient vector is formulated. To address this high-dimensional non-convex optimization problem, we propose a novel unsupervised deep neural network (DNN) based algorithm with a simple structure. Different from most previous works that adopt the iterative optimization to solve the problem, the proposed DNN based algorithm directly obtains the BS precoding and IRS phase shifts as the output of the DNN. Simulation results reveal that the proposed DNN-based algorithm outperforms the benchmark methods with regard to SKR.

This paper investigates a physical layer key generation (PLKG) scheme for reconfigurable intelligent surface (RIS)-aided millimeter-wave (mmWave) multiple-input multipleoutput (MIMO) systems. Unlike traditional PLKG which relies on channel state information (CSI), we exploit the virtual angles of departure (AoDs). To accurately estimate these angles, we propose a redundant transforming matrix-based compressive sampling matching pursuit (RTMCoSa) method. We then derive the secret key rate (SKR) of the RIS-aided mmWave system. Simulation results demonstrate that the RTMCoSa method outperforms existing orthogonal matching pursuit (OMP) methods in channel probing for angle information. What is more, the proposed key generation scheme surpasses traditional CSI-based methods in SKR when the SNR is low. And the SKR of our method maintains robust when the SNR decreases.

The one time pad (OTP) technique can provide information-theoretic security and physical layer key generation (PLKG) is a promising candidate for OTP. In the joint PLKG and OTP scheme, channel probing, information reconciliation and encrypted packet transmission all need to occupy channel time. In this paper, we propose a time slot allocation scheme for PLKG in OTP. We divide the process of OTP into three stages: channel probing, information reconciliation and encrypted packet transmission. Then, we analyze the secret key capacity and the time slot cost of the whole process. Under the premise of ensuring safety, we give a time slot reduction strategy to reduce time slot cost. Meanwhile, we propose a time slot allocation optimization algorithm to realize the strategy. Simulation results verify that our scheme can significantly reduce the time slot cost.

In quasi-static environments, the key generation rate (KGR) of wireless physical layer key generation (PLKG) can be greatly limited. In this paper, we aim to improve the KGR by reconfigurable intelligent surface (RIS) and jointly design the quantizer and phase shift matrix. We give a reconfiguration strategy able to quantify both real and imaginary parts at the same time and give its closed-form solution of the phase shift matrix. To reduce the bit disagreement ratio (BDR), we design a quantizer suitable for RIS-assisted PLKG which uses non-uniform quantization. To achieve high KGR while maintain low BDR, we analyze the BDR in different quantization levels and realize an adaptive quantization selection. Meanwhile, the simulation results verify that the KGR of the proposed scheme is more than twice that of the existing scheme under a certain BDR threshold.

Physical layer key generation (PLKG) exploits the distributed entropy source of wireless channels to generate secret keys for legitimate users. When the millimeter wave (mmWave) channel is blocked, reconfigurable intelligent surfaces (RISs) have emerged as a prospective approach to constructing reflected channels and improving the secret key rate (SKR). This paper investigates the key generation scheme for the RIS-aided mmWave system. We study the beam domain channel model and exploit the sparsity of mmWave bands to reduce the pilot overhead. We propose a channel probing method to acquire the reciprocal angular information and channel gains. To analyze the SKR, we investigate the channel covariance matrix of beam domain channels. We find that the channel gains of beams are uncorrelated which increases the randomness of secret keys. Considering an eavesdropper, we derive the analytical expressions of SKR when the eavesdropping channel has overlapping clusters with the legitimate channel. Simulations validate that the proposed PLKG scheme outperforms existing schemes.

Physical layer key generation (PLKG) is a promising technique to establish symmetric keys between resource-constrained legitimate users. However, PLKG suffers from a low key rate in harsh environments where channel randomness is limited. To address the problem, reconfigurable intelligent surfaces (RISs) are introduced to reshape the channels by controlling massive reflecting elements, which can provide more channel diversity. In this paper, we design a channel probing protocol to fully extract the randomness from the cascaded channel, i.e., the channels through reflecting elements. We derive the analytical expressions of the key rate and design a water-filling algorithm based on the Karush-Kuhn-Tucker (KKT) conditions to find the upper bound. To find the optimal precoding and phase shift matrices, we propose an algorithm based on the Grassmann manifold optimization methods. The system is evaluated in terms of the key rate, bit disagreement rate (BDR) and randomness. Simulation results show that our protocols significantly improve the key rate as compared to existing protocols. Compared to multiple-antennas systems without a RIS, our proposed method achieves an average 9.51 dB performance gain when the side length of an element is 1/4 wavelength and the Rician factor is 0 dB.

The provably NP-hard problem of finding optimal piecewise linear approximation for images is extended from 1D curve fitting to 2D surface fitting by a dual-agent algorithm. The results not only yield a storage-efficient codec for range, or intensity, images but also a surface triangulation technique to generate a succinct, accurate and visually pleasant 3D visualization model. Compared with the traditional piecewise linear image coding (PLIC) algorithms, triangulation of a range image is more adaptive due to conformity of the shape, orientation and size of triangles with the image contents. The triangularization algorithm presented here differs from previous approaches in that it strives to minimize the total number of triangles (or vertices) needed to approximate the image surface while keeping the deviation of any intensity value to within a prescribed error tolerance. Unlike most methods of bottom-up triangularization, which could bog down before any mesh simplification even begins, this algorithm fits the surface in a top-down manner, avoiding the generation of most unnecessary triangles. When combined with an efficient 3D triangulation-encoding scheme, the algorithm achieves compact code length with guaranteed error bound, thus providing a more faithful representation of all image features. A variety of benchmark test images have been experimented with and compared.