New submissions for Wed, 3 Jan 24

[zoq]

Keyword: sgd

There is no result

Keyword: optimization

TrailBlazer: Trajectory Control for Diffusion-Based Video Generation

• Authors: Authors: Wan-Duo Kurt Ma, J.P. Lewis, W. Bastiaan Kleijn
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2401.00896
• Pdf link: https://arxiv.org/pdf/2401.00896
• Abstract Within recent approaches to text-to-video (T2V) generation, achieving controllability in the synthesized video is often a challenge. Typically, this issue is addressed by providing low-level per-frame guidance in the form of edge maps, depth maps, or an existing video to be altered. However, the process of obtaining such guidance can be labor-intensive. This paper focuses on enhancing controllability in video synthesis by employing straightforward bounding boxes to guide the subject in various ways, all without the need for neural network training, finetuning, optimization at inference time, or the use of pre-existing videos. Our algorithm, TrailBlazer, is constructed upon a pre-trained (T2V) model, and easy to implement. The subject is directed by a bounding box through the proposed spatial and temporal attention map editing. Moreover, we introduce the concept of keyframing, allowing the subject trajectory and overall appearance to be guided by both a moving bounding box and corresponding prompts, without the need to provide a detailed mask. The method is efficient, with negligible additional computation relative to the underlying pre-trained model. Despite the simplicity of the bounding box guidance, the resulting motion is surprisingly natural, with emergent effects including perspective and movement toward the virtual camera as the box size increases.

  Unsupervised Graph-based Learning Method for Sub-band Allocation in 6G Subnetworks

• Authors: Authors: Daniel Abode, Ramoni Adeogun, Lou Salaün, Renato Abreu, Thomas Jacobsen, Gilberto Berardinelli
• Subjects: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG); Systems and Control (eess.SY)
• Arxiv link: https://arxiv.org/abs/2401.00950
• Pdf link: https://arxiv.org/pdf/2401.00950
• Abstract In this paper, we present an unsupervised approach for frequency sub-band allocation in wireless networks using graph-based learning. We consider a dense deployment of subnetworks in the factory environment with a limited number of sub-bands which must be optimally allocated to coordinate inter-subnetwork interference. We model the subnetwork deployment as a conflict graph and propose an unsupervised learning approach inspired by the graph colouring heuristic and the Potts model to optimize the sub-band allocation using graph neural networks. The numerical evaluation shows that the proposed method achieves close performance to the centralized greedy colouring sub-band allocation heuristic with lower computational time complexity. In addition, it incurs reduced signalling overhead compared to iterative optimization heuristics that require all the mutual interfering channel information. We further demonstrate that the method is robust to different network settings.

  Nature-Inspired Algorithms in Optimization: Introduction, Hybridization and Insights

• Authors: Authors: Xin-She Yang
• Subjects: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Optimization and Control (math.OC)
• Arxiv link: https://arxiv.org/abs/2401.00976
• Pdf link: https://arxiv.org/pdf/2401.00976
• Abstract Many problems in science and engineering are optimization problems, which may require sophisticated optimization techniques to solve. Nature-inspired algorithms are a class of metaheuristic algorithms for optimization, and some algorithms or variants are often developed by hybridization. Benchmarking is also important in evaluating the performance of optimization algorithms. This chapter focuses on the overview of optimization, nature-inspired algorithms and the role of hybridization. We will also highlight some issues with hybridization of algorithms.

  Evolutionary Alternating Direction Method of Multipliers for Constrained Multi-Objective Optimization with Unknown Constraints

• Authors: Authors: Shuang Li, Ke Li, Wei Li, Ming Yang
• Subjects: Neural and Evolutionary Computing (cs.NE)
• Arxiv link: https://arxiv.org/abs/2401.00978
• Pdf link: https://arxiv.org/pdf/2401.00978
• Abstract Constrained multi-objective optimization problems (CMOPs) pervade real-world applications in science, engineering, and design. Constraint violation has been a building block in designing evolutionary multi-objective optimization algorithms for solving constrained multi-objective optimization problems. However, in certain scenarios, constraint functions might be unknown or inadequately defined, making constraint violation unattainable and potentially misleading for conventional constrained evolutionary multi-objective optimization algorithms. To address this issue, we present the first of its kind evolutionary optimization framework, inspired by the principles of the alternating direction method of multipliers that decouples objective and constraint functions. This framework tackles CMOPs with unknown constraints by reformulating the original problem into an additive form of two subproblems, each of which is allotted a dedicated evolutionary population. Notably, these two populations operate towards complementary evolutionary directions during their optimization processes. In order to minimize discrepancy, their evolutionary directions alternate, aiding the discovery of feasible solutions. Comparative experiments conducted against five state-of-the-art constrained evolutionary multi-objective optimization algorithms, on 120 benchmark test problem instances with varying properties, as well as two real-world engineering optimization problems, demonstrate the effectiveness and superiority of our proposed framework. Its salient features include faster convergence and enhanced resilience to various Pareto front shapes.

  Real-Time Object Detection in Occluded Environment with Background Cluttering Effects Using Deep Learning

• Authors: Authors: Syed Muhammad Aamir, Hongbin Ma, Malak Abid Ali Khan, Muhammad Aaqib
• Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
• Arxiv link: https://arxiv.org/abs/2401.00986
• Pdf link: https://arxiv.org/pdf/2401.00986
• Abstract Detection of small, undetermined moving objects or objects in an occluded environment with a cluttered background is the main problem of computer vision. This greatly affects the detection accuracy of deep learning models. To overcome these problems, we concentrate on deep learning models for real-time detection of cars and tanks in an occluded environment with a cluttered background employing SSD and YOLO algorithms and improved precision of detection and reduce problems faced by these models. The developed method makes the custom dataset and employs a preprocessing technique to clean the noisy dataset. For training the developed model we apply the data augmentation technique to balance and diversify the data. We fine-tuned, trained, and evaluated these models on the established dataset by applying these techniques and highlighting the results we got more accurately than without applying these techniques. The accuracy and frame per second of the SSD-Mobilenet v2 model are higher than YOLO V3 and YOLO V4. Furthermore, by employing various techniques like data enhancement, noise reduction, parameter optimization, and model fusion we improve the effectiveness of detection and recognition. We further added a counting algorithm, and target attributes experimental comparison, and made a graphical user interface system for the developed model with features of object counting, alerts, status, resolution, and frame per second. Subsequently, to justify the importance of the developed method analysis of YOLO V3, V4, and SSD were incorporated. Which resulted in the overall completion of the proposed method.

  Safety and Performance, Why Not Both? Bi-Objective Optimized Model Compression against Heterogeneous Attacks Toward AI Software Deployment

• Authors: Authors: Jie Zhu, Leye Wang, Xiao Han, Anmin Liu, Tao Xie
• Subjects: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Software Engineering (cs.SE)
• Arxiv link: https://arxiv.org/abs/2401.00996
• Pdf link: https://arxiv.org/pdf/2401.00996
• Abstract The size of deep learning models in artificial intelligence (AI) software is increasing rapidly, hindering the large-scale deployment on resource-restricted devices (e.g., smartphones). To mitigate this issue, AI software compression plays a crucial role, which aims to compress model size while keeping high performance. However, the intrinsic defects in a big model may be inherited by the compressed one. Such defects may be easily leveraged by adversaries, since a compressed model is usually deployed in a large number of devices without adequate protection. In this article, we aim to address the safe model compression problem from the perspective of safety-performance co-optimization. Specifically, inspired by the test-driven development (TDD) paradigm in software engineering, we propose a test-driven sparse training framework called SafeCompress. By simulating the attack mechanism as safety testing, SafeCompress can automatically compress a big model to a small one following the dynamic sparse training paradigm. Then, considering two kinds of representative and heterogeneous attack mechanisms, i.e., black-box membership inference attack and white-box membership inference attack, we develop two concrete instances called BMIA-SafeCompress and WMIA-SafeCompress. Further, we implement another instance called MMIA-SafeCompress by extending SafeCompress to defend against the occasion when adversaries conduct black-box and white-box membership inference attacks simultaneously. We conduct extensive experiments on five datasets for both computer vision and natural language processing tasks. The results show the effectiveness and generalizability of our framework. We also discuss how to adapt SafeCompress to other attacks besides membership inference attack, demonstrating the flexibility of SafeCompress.

  Towards Net-Zero Carbon Emissions in Network AI for 6G and Beyond

• Authors: Authors: Peng Zhang, Yong Xiao, Yingyu Li, Xiaohu Ge, Guangming Shi, Yang Yang
• Subjects: Networking and Internet Architecture (cs.NI); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
• Arxiv link: https://arxiv.org/abs/2401.01007
• Pdf link: https://arxiv.org/pdf/2401.01007
• Abstract A global effort has been initiated to reduce the worldwide greenhouse gas (GHG) emissions, primarily carbon emissions, by half by 2030 and reach net-zero by 2050. The development of 6G must also be compliant with this goal. Unfortunately, developing a sustainable and net-zero emission systems to meet the users' fast growing demands on mobile services, especially smart services and applications, may be much more challenging than expected. Particularly, despite the energy efficiency improvement in both hardware and software designs, the overall energy consumption and carbon emission of mobile networks are still increasing at a tremendous speed. The growing penetration of resource-demanding AI algorithms and solutions further exacerbate this challenge. In this article, we identify the major emission sources and introduce an evaluation framework for analyzing the lifecycle of network AI implementations. A novel joint dynamic energy trading and task allocation optimization framework, called DETA, has been introduced to reduce the overall carbon emissions. We consider a federated edge intelligence-based network AI system as a case study to verify the effectiveness of our proposed solution. Experimental results based on a hardware prototype suggest that our proposed solution can reduce carbon emissions of network AI systems by up to 74.9%. Finally, open problems and future directions are discussed.

  Quantum State Preparation Using an Exact CNOT Synthesis Formulation

• Authors: Authors: Hanyu Wang, Bochen Tan, Jason Cong, Giovanni De Micheli
• Subjects: Information Theory (cs.IT)
• Arxiv link: https://arxiv.org/abs/2401.01009
• Pdf link: https://arxiv.org/pdf/2401.01009
• Abstract Minimizing the use of CNOT gates in quantum state preparation is a crucial step in quantum compilation, as they introduce coupling constraints and more noise than single-qubit gates. Reducing the number of CNOT gates can lead to more efficient and accurate quantum computations. However, the lack of compatibility to model superposition and entanglement challenges the scalability and optimality of CNOT optimization algorithms on classical computers. In this paper, we propose an effective state preparation algorithm using an exact CNOT synthesis formulation. Our method represents a milestone as the first design automation algorithm to surpass manual design, reducing the best CNOT numbers to prepare a Dicke state by 2x. For general states with up to 20 qubits, our method reduces the CNOT number by 9% and 32% for dense and sparse states, on average, compared to the latest algorithms.

  Elastic Multi-Gradient Descent for Parallel Continual Learning

• Authors: Authors: Fan Lyu, Wei Feng, Yuepan Li, Qing Sun, Fanhua Shang, Liang Wan, Liang Wang
• Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
• Arxiv link: https://arxiv.org/abs/2401.01054
• Pdf link: https://arxiv.org/pdf/2401.01054
• Abstract The goal of Continual Learning (CL) is to continuously learn from new data streams and accomplish the corresponding tasks. Previously studied CL assumes that data are given in sequence nose-to-tail for different tasks, thus indeed belonging to Serial Continual Learning (SCL). This paper studies the novel paradigm of Parallel Continual Learning (PCL) in dynamic multi-task scenarios, where a diverse set of tasks is encountered at different time points. PCL presents challenges due to the training of an unspecified number of tasks with varying learning progress, leading to the difficulty of guaranteeing effective model updates for all encountered tasks. In our previous conference work, we focused on measuring and reducing the discrepancy among gradients in a multi-objective optimization problem, which, however, may still contain negative transfers in every model update. To address this issue, in the dynamic multi-objective optimization problem, we introduce task-specific elastic factors to adjust the descent direction towards the Pareto front. The proposed method, called Elastic Multi-Gradient Descent (EMGD), ensures that each update follows an appropriate Pareto descent direction, minimizing any negative impact on previously learned tasks. To balance the training between old and new tasks, we also propose a memory editing mechanism guided by the gradient computed using EMGD. This editing process updates the stored data points, reducing interference in the Pareto descent direction from previous tasks. Experiments on public datasets validate the effectiveness of our EMGD in the PCL setting.

  A prediction-correction based iterative convolution-thresholding method for topology optimization of heat transfer problems

• Authors: Authors: Huangxin Chen, Piaopiao Dong, Dong Wang, Xiao-Ping Wang
• Subjects: Numerical Analysis (math.NA); Computational Engineering, Finance, and Science (cs.CE)
• Arxiv link: https://arxiv.org/abs/2401.01069
• Pdf link: https://arxiv.org/pdf/2401.01069
• Abstract In this paper, we propose an iterative convolution-thresholding method (ICTM) based on prediction-correction for solving the topology optimization problem in steady-state heat transfer equations. The problem is formulated as a constrained minimization problem of the complementary energy, incorporating a perimeter/surface-area regularization term, while satisfying a steady-state heat transfer equation. The decision variables of the optimization problem represent the domains of different materials and are represented by indicator functions. The perimeter/surface-area term of the domain is approximated using Gaussian kernel convolution with indicator functions. In each iteration, the indicator function is updated using a prediction-correction approach. The prediction step is based on the variation of the objective functional by imposing the constraints, while the correction step ensures the monotonically decreasing behavior of the objective functional. Numerical results demonstrate the efficiency and robustness of our proposed method, particularly when compared to classical approaches based on the ICTM.

  A Novel Dual-Stage Evolutionary Algorithm for Finding Robust Solutions

• Authors: Authors: Wei Du, Wenxuan Fang, Chen Liang, Yang Tang, Yaochu Jin
• Subjects: Neural and Evolutionary Computing (cs.NE)
• Arxiv link: https://arxiv.org/abs/2401.01070
• Pdf link: https://arxiv.org/pdf/2401.01070
• Abstract In robust optimization problems, the magnitude of perturbations is relatively small. Consequently, solutions within certain regions are less likely to represent the robust optima when perturbations are introduced. Hence, a more efficient search process would benefit from increased opportunities to explore promising regions where global optima or good local optima are situated. In this paper, we introduce a novel robust evolutionary algorithm named the dual-stage robust evolutionary algorithm (DREA) aimed at discovering robust solutions. DREA operates in two stages: the peak-detection stage and the robust solution-searching stage. The primary objective of the peak-detection stage is to identify peaks in the fitness landscape of the original optimization problem. Conversely, the robust solution-searching stage focuses on swiftly identifying the robust optimal solution using information obtained from the peaks discovered in the initial stage. These two stages collectively enable the proposed DREA to efficiently obtain the robust optimal solution for the optimization problem. This approach achieves a balance between solution optimality and robustness by separating the search processes for optimal and robust optimal solutions. Experimental results demonstrate that DREA significantly outperforms five state-of-the-art algorithms across 18 test problems characterized by diverse complexities. Moreover, when evaluated on higher-dimensional robust optimization problems (100-$D$ and 200-$D$), DREA also demonstrates superior performance compared to all five counterpart algorithms.

  Constrained Online Two-stage Stochastic Optimization: Algorithm with (and without) Predictions

• Authors: Authors: Piao Hu, Jiashuo Jiang, Guodong Lyu, Hao Su
• Subjects: Machine Learning (cs.LG)
• Arxiv link: https://arxiv.org/abs/2401.01077
• Pdf link: https://arxiv.org/pdf/2401.01077
• Abstract We consider an online two-stage stochastic optimization with long-term constraints over a finite horizon of $T$ periods. At each period, we take the first-stage action, observe a model parameter realization and then take the second-stage action from a feasible set that depends both on the first-stage decision and the model parameter. We aim to minimize the cumulative objective value while guaranteeing that the long-term average second-stage decision belongs to a set. We develop online algorithms for the online two-stage problem from adversarial learning algorithms. Also, the regret bound of our algorithm can be reduced to the regret bound of embedded adversarial learning algorithms. Based on this framework, we obtain new results under various settings. When the model parameters are drawn from unknown non-stationary distributions and we are given machine-learned predictions of the distributions, we develop a new algorithm from our framework with a regret $O(W_T+\sqrt{T})$, where $W_T$ measures the total inaccuracy of the machine-learned predictions. We then develop another algorithm that works when no machine-learned predictions are given and show the performances.

  Imperio: Language-Guided Backdoor Attacks for Arbitrary Model Control

• Authors: Authors: Ka-Ho Chow, Wenqi Wei, Lei Yu
• Subjects: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
• Arxiv link: https://arxiv.org/abs/2401.01085
• Pdf link: https://arxiv.org/pdf/2401.01085
• Abstract Revolutionized by the transformer architecture, natural language processing (NLP) has received unprecedented attention. While advancements in NLP models have led to extensive research into their backdoor vulnerabilities, the potential for these advancements to introduce new backdoor threats remains unexplored. This paper proposes Imperio, which harnesses the language understanding capabilities of NLP models to enrich backdoor attacks. Imperio provides a new model control experience. It empowers the adversary to control the victim model with arbitrary output through language-guided instructions. This is achieved using a language model to fuel a conditional trigger generator, with optimizations designed to extend its language understanding capabilities to backdoor instruction interpretation and execution. Our experiments across three datasets, five attacks, and nine defenses confirm Imperio's effectiveness. It can produce contextually adaptive triggers from text descriptions and control the victim model with desired outputs, even in scenarios not encountered during training. The attack maintains a high success rate across complex datasets without compromising the accuracy of clean inputs and also exhibits resilience against representative defenses. The source code is available at \url{https://khchow.com/Imperio}.

  Q-Refine: A Perceptual Quality Refiner for AI-Generated Image

• Authors: Authors: Chunyi Li, Haoning Wu, Zicheng Zhang, Hongkun Hao, Kaiwei Zhang, Lei Bai, Xiaohong Liu, Xiongkuo Min, Weisi Lin, Guangtao Zhai
• Subjects: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
• Arxiv link: https://arxiv.org/abs/2401.01117
• Pdf link: https://arxiv.org/pdf/2401.01117
• Abstract With the rapid evolution of the Text-to-Image (T2I) model in recent years, their unsatisfactory generation result has become a challenge. However, uniformly refining AI-Generated Images (AIGIs) of different qualities not only limited optimization capabilities for low-quality AIGIs but also brought negative optimization to high-quality AIGIs. To address this issue, a quality-award refiner named Q-Refine is proposed. Based on the preference of the Human Visual System (HVS), Q-Refine uses the Image Quality Assessment (IQA) metric to guide the refining process for the first time, and modify images of different qualities through three adaptive pipelines. Experimental shows that for mainstream T2I models, Q-Refine can perform effective optimization to AIGIs of different qualities. It can be a general refiner to optimize AIGIs from both fidelity and aesthetic quality levels, thus expanding the application of the T2I generation models.

  Wireless 6G Connectivity for Massive Number of Devices and Critical Services

• Authors: Authors: Anders E. Kalør, Giuseppe Duris, Sinem Coleri, Stefan Parkvall, Wei Yu, Andreas Mueller, Petar Popovski
• Subjects: Information Theory (cs.IT)
• Arxiv link: https://arxiv.org/abs/2401.01127
• Pdf link: https://arxiv.org/pdf/2401.01127
• Abstract Compared to the generations up to 4G, whose main focus was on broadband and coverage aspects, 5G has expanded the scope of wireless cellular systems towards embracing two new types of connectivity: massive machine-type communication (mMTC) and ultra-reliable low-latency communications (URLLC). This paper will discuss the possible evolution of these two types of connectivity within the umbrella of 6G wireless systems. The paper consists of three parts. The first part deals with the connectivity for a massive number of devices. While mMTC research in 5G was predominantly focused on the problem of uncoordinated access in the uplink for a large number of devices, the traffic patterns in 6G may become more symmetric, leading to closed-loop massive connectivity. One of the drivers for this is distributed learning/inference. The second part of the paper will discuss the evolution of wireless connectivity for critical services. While latency and reliability are tightly coupled in 5G, 6G will support a variety of safety critical control applications with different types of timing requirements, as evidenced by the emergence of metrics related to information freshness and information value. Additionally, ensuring ultra-high reliability for safety critical control applications requires modeling and estimation of the tail statistics of the wireless channel, queue length, and delay. The fulfillment of these stringent requirements calls for the development of novel AI-based techniques, incorporating optimization theory, explainable AI, generative AI and digital twins. The third part will analyze the coexistence of massive connectivity and critical services. We will consider scenarios in which a massive number of devices need to support traffic patterns of mixed criticality. This will be followed by a discussion about the management of wireless resources shared by services with different criticality.

  A Stochastic-MILP dispatch optimization model for Concentrated Solar Thermal under uncertainty

• Authors: Authors: Navid Mohammadzadeh, Huy Truong-Ba, Michael E. Cholette, Theodore A. Steinberg, Giampaolo Manzolini
• Subjects: Systems and Control (eess.SY)
• Arxiv link: https://arxiv.org/abs/2401.01133
• Pdf link: https://arxiv.org/pdf/2401.01133
• Abstract Concentrated Solar Thermal (CST) offers a promising solution for large-scale solar energy utilization as Thermal Energy Storage (TES) enables electricity generation independently of daily solar fluctuations, shifting to high-priced electricity intervals. The development of dispatch planning tools is mandatory to account for uncertainties associated with solar irradiation and electricity price forecasts as well as limited storage capacity. This study proposes the Stochastic Mixed Integer Linear Program (SMILP) to maximize expected profit within a specified scenario space. The SMILP scenario space is generated by different Empirical Cumulative Distribution Function percentiles of the potential solar energy to accumulate in storage and the expected profit is estimated using the Sample Average Approximation (SAA) method. SMILP exhibits robust performance, however, its computational time poses a challenge. Thus, three heuristic solutions are developed which run a set of deterministic optimizations on different historical weather profiles to generate candidate dispatching plans (DPs). The candidate DP with the best average performance on all profiles is then selected. The new methods were applied to a case study for a 115 MW CST plant in South Australia. When the historical database has a limited set of historical weather profiles, the SMILP achieves 6% to 9% higher profit than the closest benchmark when the DP is applied to novel weather conditions. With a large historical weather data, the performance of SMILP and Heuristic-2 becomes nearly identical because the SMILP can only utilize a limited number of trajectories for optimization without becoming computationally infeasible. In this case, Heuristic-2 emerges a practical alternative, since it provides similar average profit in a reasonable time (saving about 7 hours in computing time).

  Joint Offloading and Resource Allocation for Hybrid Cloud and Edge Computing in SAGINs: A Decision Assisted Hybrid Action Space Deep Reinforcement Learning Approach

• Authors: Authors: Chong Huang, Gaojie Chen, Pei Xiao, Yue Xiao, Zhu Han, Jonathon A. Chambers
• Subjects: Information Theory (cs.IT); Distributed, Parallel, and Cluster Computing (cs.DC)
• Arxiv link: https://arxiv.org/abs/2401.01140
• Pdf link: https://arxiv.org/pdf/2401.01140
• Abstract In recent years, the amalgamation of satellite communications and aerial platforms into space-air-ground integrated network (SAGINs) has emerged as an indispensable area of research for future communications due to the global coverage capacity of low Earth orbit (LEO) satellites and the flexible Deployment of aerial platforms. This paper presents a deep reinforcement learning (DRL)-based approach for the joint optimization of offloading and resource allocation in hybrid cloud and multi-access edge computing (MEC) scenarios within SAGINs. The proposed system considers the presence of multiple satellites, clouds and unmanned aerial vehicles (UAVs). The multiple tasks from ground users are modeled as directed acyclic graphs (DAGs). With the goal of reducing energy consumption and latency in MEC, we propose a novel multi-agent algorithm based on DRL that optimizes both the offloading strategy and the allocation of resources in the MEC infrastructure within SAGIN. A hybrid action algorithm is utilized to address the challenge of hybrid continuous and discrete action space in the proposed problems, and a decision-assisted DRL method is adopted to reduce the impact of unavailable actions in the training process of DRL. Through extensive simulations, the results demonstrate the efficacy of the proposed learning-based scheme, the proposed approach consistently outperforms benchmark schemes, highlighting its superior performance and potential for practical applications.

  Enhancing Communication Efficiency of Semantic Transmission via Joint Processing Technique

• Authors: Authors: Xumin Pu, Tiantian Lei, Wanli Wen, Qianbin Chen
• Subjects: Information Theory (cs.IT)
• Arxiv link: https://arxiv.org/abs/2401.01143
• Pdf link: https://arxiv.org/pdf/2401.01143
• Abstract This work presents a novel semantic transmission framework in wireless networks, leveraging the joint processing technique. Our framework enables multiple cooperating base stations to efficiently transmit semantic information to multiple users simultaneously. To enhance the semantic communication efficiency of the transmission framework, we formulate an optimization problem with the objective of maximizing the semantic spectral efficiency of the framework and propose a lowcomplexity dynamic semantic mapping and resource allocation algorithm. This algorithm, based on deep reinforcement learning and alternative optimization, achieves near-optimal performance while reducing computational complexity. Simulation results validate the effectiveness of the proposed algorithm, bridging the research gap and facilitating the practical implementation of semantic communication systems.

  En3D: An Enhanced Generative Model for Sculpting 3D Humans from 2D Synthetic Data

• Authors: Authors: Yifang Men, Biwen Lei, Yuan Yao, Miaomiao Cui, Zhouhui Lian, Xuansong Xie
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2401.01173
• Pdf link: https://arxiv.org/pdf/2401.01173
• Abstract We present En3D, an enhanced generative scheme for sculpting high-quality 3D human avatars. Unlike previous works that rely on scarce 3D datasets or limited 2D collections with imbalanced viewing angles and imprecise pose priors, our approach aims to develop a zero-shot 3D generative scheme capable of producing visually realistic, geometrically accurate and content-wise diverse 3D humans without relying on pre-existing 3D or 2D assets. To address this challenge, we introduce a meticulously crafted workflow that implements accurate physical modeling to learn the enhanced 3D generative model from synthetic 2D data. During inference, we integrate optimization modules to bridge the gap between realistic appearances and coarse 3D shapes. Specifically, En3D comprises three modules: a 3D generator that accurately models generalizable 3D humans with realistic appearance from synthesized balanced, diverse, and structured human images; a geometry sculptor that enhances shape quality using multi-view normal constraints for intricate human anatomy; and a texturing module that disentangles explicit texture maps with fidelity and editability, leveraging semantical UV partitioning and a differentiable rasterizer. Experimental results show that our approach significantly outperforms prior works in terms of image quality, geometry accuracy and content diversity. We also showcase the applicability of our generated avatars for animation and editing, as well as the scalability of our approach for content-style free adaptation.

  Deep-ELA: Deep Exploratory Landscape Analysis with Self-Supervised Pretrained Transformers for Single- and Multi-Objective Continuous Optimization Problems

• Authors: Authors: Moritz Vinzent Seiler, Pascal Kerschke, Heike Trautmann
• Subjects: Machine Learning (cs.LG)
• Arxiv link: https://arxiv.org/abs/2401.01192
• Pdf link: https://arxiv.org/pdf/2401.01192
• Abstract In many recent works, the potential of Exploratory Landscape Analysis (ELA) features to numerically characterize, in particular, single-objective continuous optimization problems has been demonstrated. These numerical features provide the input for all kinds of machine learning tasks on continuous optimization problems, ranging, i.a., from High-level Property Prediction to Automated Algorithm Selection and Automated Algorithm Configuration. Without ELA features, analyzing and understanding the characteristics of single-objective continuous optimization problems would be impossible. Yet, despite their undisputed usefulness, ELA features suffer from several drawbacks. These include, in particular, (1.) a strong correlation between multiple features, as well as (2.) its very limited applicability to multi-objective continuous optimization problems. As a remedy, recent works proposed deep learning-based approaches as alternatives to ELA. In these works, e.g., point-cloud transformers were used to characterize an optimization problem's fitness landscape. However, these approaches require a large amount of labeled training data. Within this work, we propose a hybrid approach, Deep-ELA, which combines (the benefits of) deep learning and ELA features. Specifically, we pre-trained four transformers on millions of randomly generated optimization problems to learn deep representations of the landscapes of continuous single- and multi-objective optimization problems. Our proposed framework can either be used out-of-the-box for analyzing single- and multi-objective continuous optimization problems, or subsequently fine-tuned to various tasks focussing on algorithm behavior and problem understanding.

  KCES: A Workflow Containerization Scheduling Scheme Under Cloud-Edge Collaboration Framework

• Authors: Authors: Chenggang Shan, Runze Gao, Qinghua Han, Zhen Yang, Jinhui Zhang, Yuanqing Xia
• Subjects: Distributed, Parallel, and Cluster Computing (cs.DC)
• Arxiv link: https://arxiv.org/abs/2401.01217
• Pdf link: https://arxiv.org/pdf/2401.01217
• Abstract As more IoT applications gradually move towards the cloud-edge collaborative mode, the containerized scheduling of workflows extends from the cloud to the edge. However, given the high delay of the communication network, loose coupling of structure, and resource heterogeneity between cloud and edge, workflow containerization scheduling in the cloud-edge scenarios faces the difficulty of resource coordination and application collaboration management. To address these two issues, we propose a KubeEdge-Cloud-Edge-Scheduling scheme named KCES, a workflow containerization scheduling scheme for the KubeEdge cloud-edge framework. The KCES includes a cloud-edge workflow scheduling engine for KubeEdge and workflow scheduling strategies for task horizontal roaming and vertical offloading. Considering the scheduling optimization of cloud-edge workflows, this paper proposes a cloud-edge workflow scheduling model and cloud-edge node model and designs a cloud-edge workflow scheduling engine to maximize cloud-edge resource utilization under the constraint of workflow task delay. A cloud-edge resource hybrid management technology is used to design the cloud-edge resource evaluation and resource allocation algorithms to achieve cloud-edge resource collaboration. Based on the ideas of distributed functional roles and the hierarchical division of computing power, the horizontal roaming among the edges and vertical offloading strategies between the cloud and edges for workflow tasks are designed to realize the cloud-edge application collaboration. Through a customized IoT application workflow instance, experimental results show that KCES is superior to the baseline in total workflow duration, average workflow duration, and resource usage and has the capabilities of horizontal roaming and vertical offloading for workflow tasks.

  Optimal Synthesis of Finite State Machines with Universal Gates using Evolutionary Algorithm

• Authors: Authors: Noor Ullah, Khawaja M. Yahya, Irfan Ahmed
• Subjects: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI)
• Arxiv link: https://arxiv.org/abs/2401.01265
• Pdf link: https://arxiv.org/pdf/2401.01265
• Abstract This work presents an optimization method for the synthesis of finite state machines. The focus is on the reduction in the on-chip area and the cost of the circuit. A list of finite state machines from MCNC91 benchmark circuits have been evolved using Cartesian Genetic Programming. On the average, almost 30% of reduction in the total number of gates has been achieved. The effects of some parameters on the evolutionary process have also been discussed in the paper.

  $f$-Divergence Based Classification: Beyond the Use of Cross-Entropy

• Authors: Authors: Nicola Novello, Andrea M. Tonello
• Subjects: Machine Learning (cs.LG); Signal Processing (eess.SP)
• Arxiv link: https://arxiv.org/abs/2401.01268
• Pdf link: https://arxiv.org/pdf/2401.01268
• Abstract In deep learning, classification tasks are formalized as optimization problems solved via the minimization of the cross-entropy. However, recent advancements in the design of objective functions allow the $f$-divergence measure to generalize the formulation of the optimization problem for classification. With this goal in mind, we adopt a Bayesian perspective and formulate the classification task as a maximum a posteriori probability problem. We propose a class of objective functions based on the variational representation of the $f$-divergence, from which we extract a list of five posterior probability estimators leveraging well-known $f$-divergences. In addition, driven by the challenge of improving the state-of-the-art approach, we propose a bottom-up method that leads us to the formulation of a new objective function (and posterior probability estimator) corresponding to a novel $f$-divergence referred to as shifted log (SL). First, we theoretically prove the convergence property of the posterior probability estimators. Then, we numerically test the set of proposed objective functions in three application scenarios: toy examples, image data sets, and signal detection/decoding problems. The analyzed tasks demonstrate the effectiveness of the proposed estimators and that the SL divergence achieves the highest classification accuracy in almost all the scenarios.

  Learning-based agricultural management in partially observable environments subject to climate variability

• Authors: Authors: Zhaoan Wang, Shaoping Xiao, Junchao Li, Jun Wang
• Subjects: Machine Learning (cs.LG)
• Arxiv link: https://arxiv.org/abs/2401.01273
• Pdf link: https://arxiv.org/pdf/2401.01273
• Abstract Agricultural management, with a particular focus on fertilization strategies, holds a central role in shaping crop yield, economic profitability, and environmental sustainability. While conventional guidelines offer valuable insights, their efficacy diminishes when confronted with extreme weather conditions, such as heatwaves and droughts. In this study, we introduce an innovative framework that integrates Deep Reinforcement Learning (DRL) with Recurrent Neural Networks (RNNs). Leveraging the Gym-DSSAT simulator, we train an intelligent agent to master optimal nitrogen fertilization management. Through a series of simulation experiments conducted on corn crops in Iowa, we compare Partially Observable Markov Decision Process (POMDP) models with Markov Decision Process (MDP) models. Our research underscores the advantages of utilizing sequential observations in developing more efficient nitrogen input policies. Additionally, we explore the impact of climate variability, particularly during extreme weather events, on agricultural outcomes and management. Our findings demonstrate the adaptability of fertilization policies to varying climate conditions. Notably, a fixed policy exhibits resilience in the face of minor climate fluctuations, leading to commendable corn yields, cost-effectiveness, and environmental conservation. However, our study illuminates the need for agent retraining to acquire new optimal policies under extreme weather events. This research charts a promising course toward adaptable fertilization strategies that can seamlessly align with dynamic climate scenarios, ultimately contributing to the optimization of crop management practices.

  Quality and Quantity of Machine Translation References for Automated Metrics

• Authors: Authors: Vilém Zouhar, Ondřej Bojar
• Subjects: Computation and Language (cs.CL)
• Arxiv link: https://arxiv.org/abs/2401.01283
• Pdf link: https://arxiv.org/pdf/2401.01283
• Abstract Automatic machine translation metrics often use human translations to determine the quality system translations. Common wisdom in the field dictates that the human references should be of very high quality. However, there are no cost-benefit analyses that could be used to guide practitioners who plan to collect references for machine translation evaluation. We find that higher-quality references lead to better metric correlations with humans at the segment-level. Having up to 7 references per segment and taking their average helps all metrics. Interestingly, the references from vendors of different qualities can be mixed together and improve metric success. Higher quality references, however, cost more to create and we frame this as an optimization problem: given a specific budget, what references should be collected to maximize metric success. These findings can be used by evaluators of shared tasks when references need to be created under a certain budget.

  Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models

• Authors: Authors: Zixiang Chen, Yihe Deng, Huizhuo Yuan, Kaixuan Ji, Quanquan Gu
• Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (stat.ML)
• Arxiv link: https://arxiv.org/abs/2401.01335
• Pdf link: https://arxiv.org/pdf/2401.01335
• Abstract Harnessing the power of human-annotated data through Supervised Fine-Tuning (SFT) is pivotal for advancing Large Language Models (LLMs). In this paper, we delve into the prospect of growing a strong LLM out of a weak one without the need for acquiring additional human-annotated data. We propose a new fine-tuning method called Self-Play fIne-tuNing (SPIN), which starts from a supervised fine-tuned model. At the heart of SPIN lies a self-play mechanism, where the LLM refines its capability by playing against instances of itself. More specifically, the LLM generates its own training data from its previous iterations, refining its policy by discerning these self-generated responses from those obtained from human-annotated data. Our method progressively elevates the LLM from a nascent model to a formidable one, unlocking the full potential of human-annotated demonstration data for SFT. Theoretically, we prove that the global optimum to the training objective function of our method is achieved only when the LLM policy aligns with the target data distribution. Empirically, we evaluate our method on several benchmark datasets including the HuggingFace Open LLM Leaderboard, MT-Bench, and datasets from Big-Bench. Our results show that SPIN can significantly improve the LLM's performance across a variety of benchmarks and even outperform models trained through direct preference optimization (DPO) supplemented with extra GPT-4 preference data. This sheds light on the promise of self-play, enabling the achievement of human-level performance in LLMs without the need for expert opponents.

  Keyword: adam

  There is no result

  Keyword: gradient

  A Bayesian Unification of Self-Supervised Clustering and Energy-Based Models

• Authors: Authors: Emanuele Sansone, Robin Manhaeve
• Subjects: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2401.00873
• Pdf link: https://arxiv.org/pdf/2401.00873
• Abstract Self-supervised learning is a popular and powerful method for utilizing large amounts of unlabeled data, for which a wide variety of training objectives have been proposed in the literature. In this study, we perform a Bayesian analysis of state-of-the-art self-supervised learning objectives, elucidating the underlying probabilistic graphical models in each class and presenting a standardized methodology for their derivation from first principles. The analysis also indicates a natural means of integrating self-supervised learning with likelihood-based generative models. We instantiate this concept within the realm of cluster-based self-supervised learning and energy models, introducing a novel lower bound which is proven to reliably penalize the most important failure modes. Furthermore, this newly proposed lower bound enables the training of a standard backbone architecture without the necessity for asymmetric elements such as stop gradients, momentum encoders, or specialized clustering layers - typically introduced to avoid learning trivial solutions. Our theoretical findings are substantiated through experiments on synthetic and real-world data, including SVHN, CIFAR10, and CIFAR100, thus showing that our objective function allows to outperform existing self-supervised learning strategies in terms of clustering, generation and out-of-distribution detection performance by a wide margin. We also demonstrate that GEDI can be integrated into a neural-symbolic framework to mitigate the reasoning shortcut problem and to learn higher quality symbolic representations thanks to the enhanced classification performance.

  Elastic Multi-Gradient Descent for Parallel Continual Learning

• Authors: Authors: Fan Lyu, Wei Feng, Yuepan Li, Qing Sun, Fanhua Shang, Liang Wan, Liang Wang
• Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
• Arxiv link: https://arxiv.org/abs/2401.01054
• Pdf link: https://arxiv.org/pdf/2401.01054
• Abstract The goal of Continual Learning (CL) is to continuously learn from new data streams and accomplish the corresponding tasks. Previously studied CL assumes that data are given in sequence nose-to-tail for different tasks, thus indeed belonging to Serial Continual Learning (SCL). This paper studies the novel paradigm of Parallel Continual Learning (PCL) in dynamic multi-task scenarios, where a diverse set of tasks is encountered at different time points. PCL presents challenges due to the training of an unspecified number of tasks with varying learning progress, leading to the difficulty of guaranteeing effective model updates for all encountered tasks. In our previous conference work, we focused on measuring and reducing the discrepancy among gradients in a multi-objective optimization problem, which, however, may still contain negative transfers in every model update. To address this issue, in the dynamic multi-objective optimization problem, we introduce task-specific elastic factors to adjust the descent direction towards the Pareto front. The proposed method, called Elastic Multi-Gradient Descent (EMGD), ensures that each update follows an appropriate Pareto descent direction, minimizing any negative impact on previously learned tasks. To balance the training between old and new tasks, we also propose a memory editing mechanism guided by the gradient computed using EMGD. This editing process updates the stored data points, reducing interference in the Pareto descent direction from previous tasks. Experiments on public datasets validate the effectiveness of our EMGD in the PCL setting.

  Global Convergence of Natural Policy Gradient with Hessian-aided Momentum Variance Reduction

• Authors: Authors: Jie Feng, Ke Wei, Jinchi Chen
• Subjects: Machine Learning (cs.LG); Optimization and Control (math.OC)
• Arxiv link: https://arxiv.org/abs/2401.01084
• Pdf link: https://arxiv.org/pdf/2401.01084
• Abstract Natural policy gradient (NPG) and its variants are widely-used policy search methods in reinforcement learning. Inspired by prior work, a new NPG variant coined NPG-HM is developed in this paper, which utilizes the Hessian-aided momentum technique for variance reduction, while the sub-problem is solved via the stochastic gradient descent method. It is shown that NPG-HM can achieve the global last iterate $\epsilon$-optimality with a sample complexity of $\mathcal{O}(\epsilon^{-2})$, which is the best known result for natural policy gradient type methods under the generic Fisher non-degenerate policy parameterizations. The convergence analysis is built upon a relaxed weak gradient dominance property tailored for NPG under the compatible function approximation framework, as well as a neat way to decompose the error when handling the sub-problem. Moreover, numerical experiments on Mujoco-based environments demonstrate the superior performance of NPG-HM over other state-of-the-art policy gradient methods.

  Learning Surface Scattering Parameters From SAR Images Using Differentiable Ray Tracing

• Authors: Authors: Jiangtao Wei, Yixiang Luomei, Xu Zhang, Feng Xu
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2401.01175
• Pdf link: https://arxiv.org/pdf/2401.01175
• Abstract Simulating high-resolution Synthetic Aperture Radar (SAR) images in complex scenes has consistently presented a significant research challenge. The development of a microwave-domain surface scattering model and its reversibility are poised to play a pivotal role in enhancing the authenticity of SAR image simulations and facilitating the reconstruction of target parameters. Drawing inspiration from the field of computer graphics, this paper proposes a surface microwave rendering model that comprehensively considers both Specular and Diffuse contributions. The model is analytically represented by the coherent spatially varying bidirectional scattering distribution function (CSVBSDF) based on the Kirchhoff approximation (KA) and the perturbation method (SPM). And SAR imaging is achieved through the synergistic combination of ray tracing and fast mapping projection techniques. Furthermore, a differentiable ray tracing (DRT) engine based on SAR images was constructed for CSVBSDF surface scattering parameter learning. Within this SAR image simulation engine, the use of differentiable reverse ray tracing enables the rapid estimation of parameter gradients from SAR images. The effectiveness of this approach has been validated through simulations and comparisons with real SAR images. By learning the surface scattering parameters, substantial enhancements in SAR image simulation performance under various observation conditions have been demonstrated.

  Keyword: super-resolution

  Noise-NeRF: Hide Information in Neural Radiance Fields using Trainable Noise

• Authors: Authors: Qinglong Huang, Yong Liao, Yanbin Hao, Pengyuan Zhou
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2401.01216
• Pdf link: https://arxiv.org/pdf/2401.01216
• Abstract Neural radiance fields (NeRF) have been proposed as an innovative 3D representation method. While attracting lots of attention, NeRF faces critical issues such as information confidentiality and security. Steganography is a technique used to embed information in another object as a means of protecting information security. Currently, there are few related studies on NeRF steganography, facing challenges in low steganography quality, model weight damage, and a limited amount of steganographic information. This paper proposes a novel NeRF steganography method based on trainable noise: Noise-NeRF. Furthermore, we propose the Adaptive Pixel Selection strategy and Pixel Perturbation strategy to improve the steganography quality and efficiency. The extensive experiments on open-source datasets show that Noise-NeRF provides state-of-the-art performances in both steganography quality and rendering quality, as well as effectiveness in super-resolution image steganography.