In this paper, we investigate the limitations of the Vector Quantized Latent Diffusion Model (VQ-LDM) in restoration tasks. We identify a performance gap between the Vector Quantization (VQ) and Diffusion Model components, manifested as a significant discrepancy between the reconstruction quality of ground truth images processed via VQ autoregression and degraded images restored by VQ-LDM. Through experiments, we attribute this gap primarily to the lack of robustness in the mapped points of VQ within the original VQ-LDM framework. To address this issue, we propose a geometric based optimization approach. First, we introduce a simple yet effective method, termed interpolation-based latent initial state optimization, which mitigates the performance gap by replacing the original mapped points with interpolated values, supported by theoretical analysis. Here, the latent initial state refers specifically to the input of the diffusion model. Building upon this, we further propose a Chebyshev center-based latent initial state optimization, an elegant theoretical solution from a geometric perspective, that further enhances restoration performance. Our improvements consistently achieve superior results across nine benchmark datasets.

Striking an optimal balance between minimal drafting latency and high speculation accuracy to enhance the inference speed of Large Language Models remains a significant challenge in speculative decoding. In this paper, we introduce Falcon, an innovative semi-autoregressive speculative decoding framework fashioned to augment both the drafter's parallelism and output quality. Falcon incorporates the Coupled Sequential Glancing Distillation technique, which fortifies inter-token dependencies within the same block, leading to increased speculation accuracy. We offer a comprehensive theoretical analysis to illuminate the underlying mechanisms. Additionally, we introduce a Custom-Designed Decoding Tree, which permits the drafter to generate multiple tokens in a single forward pass and accommodates multiple forward passes as needed, thereby boosting the number of drafted tokens and significantly improving the overall acceptance rate. Comprehensive evaluations on benchmark datasets such as MT-Bench, HumanEval, and GSM8K demonstrate Falcon's superior acceleration capabilities. The framework achieves a lossless speedup ratio ranging from 2.91x to 3.51x when tested on the Vicuna and LLaMA2-Chat model series. These results outstrip existing speculative decoding methods for LLMs, including Eagle, Medusa, Lookahead, SPS, and PLD, while maintaining a compact drafter architecture equivalent to merely two Transformer layers.

Recent advances in the areas of Internet of Things (IoT), big data, and machine learning have contributed to the rise of a growing number of complex applications. These applications will be data-intensive, delay-sensitive, and real-time as smart devices prevail more in our daily life. Ensuring quality of service (QoS) for delay-sensitive applications is a must, and fog computing is seen as one of the primary enablers for satisfying such tight QoS requirements, as it puts compute, storage, and networking resources closer to the user. In this paper, we first introduce FOGPLAN, a framework for QoS-aware dynamic fog service provisioning (QDFSP). QDFSP concerns the dynamic deployment of application services on fog nodes, or the release of application services that have previously been deployed on fog nodes, in order to meet low latency and QoS requirements of applications while minimizing cost. FOGPLAN framework is practical and operates with no assumptions and minimal information about IoT nodes. Next, we present a possible formulation (as an optimization problem) and two efficient greedy algorithms for addressing the QDFSP at one instance of time. Finally, the FOGPLAN framework is evaluated using a simulation based on real-world traffic traces.

Network function virtualization (NFV) provides an efficient and flexible way to deploy network services in the form of service function chains (SFCs) by adopting generalized equipment. However, software-based virtualized network functions (VNFs) bring new challenge for network operators in providing service availability guarantees. Traditionally, network-level protection mechanisms are considered separately from function-level VNF backup mechanisms. However, a SFC's availability cannot be guaranteed if only network- level protection mechanisms or only function-level VNF backup mechanisms are considered. In this paper, we propose a coordinated protection mechanism that adopts both backup path protection in the network and VNF replicas at nodes to guarantee a SFC's availability. The proposed mechanism determines the number of replicas required for each VNF in the SFC, and allocates the replicas to physical nodes on the working and backup paths while maintaining ordered dependency among VNFs. Simulation results show that the proposed algorithms contribute to reducing the SFC blocking and the cost of computing resources.

We consider virtual optical network (VON) mapping with the objective of minimizing total network link cost while guaranteeing VON availability, where VON availability is supported by providing shared backup path protection for selected VON links. We develop a matrix-based approach for calculating the availability of a VON mapping with shared backup path protection. In order to efficiently evaluate the maximum availability of a VON mapping, we transform the problem to a group node-weighted Steiner tree problem and propose an efficient auxiliary-graph-based availability (AA) algorithm to find a VON mapping with high availability. Based on the availability evaluation, we propose a heuristic algorithm to map the VON, and numerical results show that our algorithms are effective in achieving high availability while reducing the total link cost and the blocking rate.

We consider virtual optical network (VON) provisioning over a flexible-grid multi-domain optical network with the objective of minimizing total network cost, including the cost of transponders, regenerators, and spectrum. We propose a three-step heuristic algorithm that addresses the issues of domain selection, topology aggregation, and routing, modulation format, and spectrum assignment (RMSA) when mapping virtual optical links onto multi-domain physical optical links. We propose a domain selection technique that attempts to minimize the number of inter- domain virtual optical links. We then suggest a topology aggregation (TA) technique to exchange intra and inter-domain information between domains, and propose a method for RMSA over the aggregated topology. Numerical results show that our heuristic approach is effective in reducing total network cost.

In this paper, we consider the problem of routing, modulation, and spectrum assignment (RMSA) for data- flow transfers in elastic optical networks. We design a two-dimensional resource model, in which each data transfer with known data size can be assigned a rectangular block of resources that spans both the spectrum and time dimensions. Furthermore, the dynamic spectral resource allocation problem in elastic optical networks is simplified to the two-dimensional rectangle packing problem. We design a three-tuple for each rectangle placement, and develop a dynamic heuristic algorithm, Best Rectangle Fit (BRF), to efficiently schedule requests while minimizing fragmentation in both spectrum and time domains. We simulate the proposed algorithm, and the results show that the proposed RMSA algorithm (BRF) can greatly decrease blocking probability and increase spectrum utilization.

This paper addresses the minimum network cost problem for survivable virtual optical network mapping in flexible bandwidth optical networks. We develop an ILP model and the LBSD (the largest bandwidth requirement (LB) of virtual links versus the shortest distance (SD)) mapping approach to minimize the network cost for a given set of VONs, and we introduce two baseline mapping approaches, named LCLC (the largest computing resources' requirement versus the largest computing resources' provisioning (LC)) and LCSD (the largest computing resources' requirement versus shortest distance) mapping approaches, for comparison. Simulation results show that LBSD can achieve network cost near the ILP solutions in a 6-node network. Also, LBSD greatly reduces the cost, the spectrum usage, and the number of regenerators compared to LCLC and LCSD in the 6-node and NSFNET networks.

Network virtualization is a promising technology for the next generation network, which is required to offer a more dynamic and flexible network infrastructure. Virtual network embedding plays a vital role in the resource allocation of network virtualization. Current virtual network embedding allocates resources in an exclusive and excessive manner. For example, the whole bandwidth amount of the virtual network's peak traffic demand is allocated to the virtual network with full availability. However, such excessive resource allocation may result in resource under-utilization, leading to high user cost and low carrier revenue. To address this problem, we propose a new dynamic resource pooling and trading mechanism. The proposed mechanism is formulated as a Stackelberg game, using bandwidth as an example of the resources. We compare the user cost and the carrier revenue under the proposed mechanism against those under the exclusive resource allocation schemes. Our results show that under certain conditions, the "win-win" situation, in which the user saves cost and the carrier increases its revenue, exists and the optimal Subgame Perfect Equilibrium (SPE) point can be found.

In this paper, we aim to minimize the total network cost in flexible-grid optical networks with multiple line rates. Besides transponder cost, regenerator cost, and shared infrastructure cost, the cost of regenerator sites is also considered. We first provide the problem definition and formulate the problem as an integer linear program (ILP). We also propose a heuristic algorithm considering both selection and placement of equipment to minimize the total network cost. Simulation results show the heuristic algorithm results in up to 28% cost saving, with no significant increase in spectrum usage.

In this paper, we study the problem of regenerator pool site selection for mixed line rate optical networks (MLR-RPSS), with the objective of minimizing the number of regenerator pool sites for a given set of requests. We first provide the problem definition of MLR-RPSS and show that the MLR-RPSS problem is NP-complete. We then present four algorithms, named Independent algorithm, Sequential algorithm, MLR-combined algorithm, and Weighted MLR-combined algorithm. The performance of the algorithms is compared via simulation and results show that the Weighted MLR-combined algorithm has better performance in most cases. Also, when network load is high, the minimum number of regenerator pool sites will approach a certain limit, and some specific nodes will be more likely to be selected as regenerator pool sites.

In this paper, we study the energy-efficient impairment-constrained regenerator placement (EIRP) problem with the objective of minimizing the total energy consumption in optical networks with mixed line rates. The destination of each path is guaranteed to receive the data correctly from the source based on the regenerator placement. We first provide the problem definition of EIRP and show that the EIRP problem is NP-complete. We then formulate the problem as a mixed integer linear program (MILP) and give results for small scale problems. Two heuristic approaches, named high line rate first (HLRF) and reroute only (RO), are presented. Numerical results show that HLRF achieves good results in both large and small scale problems, and that HLRF achieves higher energy efficiency than RO.