Diffusion-based generative models provide a powerful framework for learning to sample from a complex target distribution. The remarkable empirical success of these models applied to high-dimensional signals, including...
详细信息
One way to increase solar photovoltaic penetration in the grid is management of voltage fluctuations. This is because a photovoltaic plant cannot be interconnected to the grid if it causes voltage violations. Voltage ...
One way to increase solar photovoltaic penetration in the grid is management of voltage fluctuations. This is because a photovoltaic plant cannot be interconnected to the grid if it causes voltage violations. Voltage violation is where voltage exceeds the acceptable range. Often, grid operators request photovoltaic plant owners to regulate voltage sufficiently with expensive and space-consuming static Var compensators. Unfortunately, this sometimes makes the project less feasible. This paper argues that there are better ways to regulate voltage. We ran a simulation with a 70 MWp photovoltaic plant as an addition to the grid. Without voltage regulation, voltage violations were found to be significant. This paper found that oversizing the inverter sufficiently would remove all voltage violations without deploying a static Var compensator. This is often a cheaper and space-saving solution for voltage management. This paper argues that economics and spatial efficiency of reactive power compensation devices is key to increasing photovoltaic penetration.
We consider the open federated learning (FL) systems, where clients may join and/or leave the system during the FL process. Given the variability of the number of present clients, convergence to a fixed model cannot b...
We consider the open federated learning (FL) systems, where clients may join and/or leave the system during the FL process. Given the variability of the number of present clients, convergence to a fixed model cannot be guaranteed in open systems. Instead, we resort to a new performance metric that we term the stability of open FL systems, which quantifies the magnitude of the learned model in open systems. Under the assumption that local clients’ functions are strongly convex and smooth, we theoretically quantify the radius of stability for two FL algorithms, namely local SGD and local Adam. We observe that this radius relies on several key parameters, including the function condition number as well as the variance of the stochastic gradient. Our theoretical results are further verified by numerical simulations on synthetic data.
A computing system that is based on the internet and offers users all resources as on-demand services. Servers, storage, databases, software, and networking are among these on-demand services. Usually, the user must p...
详细信息
ISBN:
(数字)9798331527549
ISBN:
(纸本)9798331527556
A computing system that is based on the internet and offers users all resources as on-demand services. Servers, storage, databases, software, and networking are among these on-demand services. Usually, the user must pay for the use of the resources; they are not their property. The number of people using this technology is increasing dramatically as a result of these features. Although it has benefits, cloud computing poses certain security and privacy risks. The authorized user's data can be stolen easily. This current research thoroughly investigates modern strategies to these problems, highlighting identity management protocols, effective encryption and blockchain-based solutions. The article identifies research gaps, focuses attention to the inadequacies of current approaches, and recommends the possible ways to improve the efficiency, scalability and trustworthiness.
In modern cell-less wireless networks, mobility management is undergoing a significant transformation, transitioning from single-link handover management to a more adaptable multi-connectivity cluster reconfiguration ...
详细信息
ISBN:
(数字)9798350343199
ISBN:
(纸本)9798350343205
In modern cell-less wireless networks, mobility management is undergoing a significant transformation, transitioning from single-link handover management to a more adaptable multi-connectivity cluster reconfiguration approach, including often conflicting objectives like energy-efficient power allocation and satisfying varying reliability requirements. In this work, we address the challenge of dynamic clustering and power allocation for unmanned aerial vehicle (UAV) communication in wireless interference networks. Our objective encompasses meeting varying reliability demands, minimizing power consumption, and reducing the frequency of cluster reconfiguration. To achieve these objectives, we introduce a novel approach based on reinforcement learning using a masked soft actor-critic algorithm, specifically tailored for dynamic clustering and power allocation.
Two primary ways to change LLM behavior are prompting and weight updates (e.g. fine-tuning). Prompting LLMs is simple and effective, specifying the desired changes explicitly in natural language, whereas weight update...
详细信息
Two primary ways to change LLM behavior are prompting and weight updates (e.g. fine-tuning). Prompting LLMs is simple and effective, specifying the desired changes explicitly in natural language, whereas weight updates provide more expressive and permanent behavior changes, specified implicitly via training on large datasets. We present a technique for "baking" prompts into the weights of an LLM. Prompt Baking converts a prompt u and initial weights θ to a new set of weights θu such that the LLM with weights θu behaves like the LLM with weights θ and prompt u. Mathematically, we minimize the KL divergence between Pθ(·|u) and Pθu(·), where P is the LLM’s probability distribution over token sequences. Across all our experiments, we find prompts can be readily baked into weight updates, often in as little as 5 minutes. Baking chain-of-thought prompts improves zero-shot performance on GSM8K, ASDiv, MBPP, ARC-Easy, ARC-Challenge, and CommonsenseQA benchmarks. Baking news headlines directly updates an LLM’s knowledge. And baking instructions & personas alleviates "prompt forgetting" over long sequences, as measured on a Persona Drift benchmark. Furthermore, stopping baking early creates "half-baked" models, allowing for continuous scaling of prompt strength. Baked models retain their sensitivity to further prompting and baking, including re-prompting with the prompt already baked in – thus amplifying the prompt’s strength. Surprisingly, the re-prompted models yield further performance gains in instruction following, as well as math reasoning and coding benchmarks (GSM8K, ASDiv, and MBPP). Taking re-prompting and re-baking to the limit yields a form of iterative self-improvement we call Prompt Pursuit, and preliminary results on instruction following exhibit dramatic performance gains with this technique. Finally, we discuss implications for AI safety, continuous model updating, improving LLM recency, enhancing real-time learning capabilities in LLM-based agents, and metho
In this study, we consider a new approach to the enhancement of classic Choquet integral as a vehicle in the processes of aggregation of classifiers o r i nformation fusion. The improvement of classification result o ...
In this study, we consider a new approach to the enhancement of classic Choquet integral as a vehicle in the processes of aggregation of classifiers o r i nformation fusion. The improvement of classification result o n a b asis o f classifier ensambles is one of the most important tasks of machine learning research community. In the previous series of works, we have introduced a conception of building Choquet-like aggregation operator using the idea inspired by one of the most common numerical methods, namely quadratures. Here, we extend this technique by using the concept which we call smoothing. We use this term to express the idea of smoothing the function under the integral symbol, and thus triggering processes that increase the elasticity of the Choquet integral. In a series of numerical experiments with anomaly detection problem, we show that the new approach is better than the existing ones in terms of accuracy and f1 score.
Atomic force microscope (AFM) is used to scan the topography of samples using a sharp probe;however, conventional AFM devices are limited by XY scanners based on piezoelectric actuation. This paper proposes a novel AF...
详细信息
In realistic distributed optimization scenarios, individual nodes possess only partial information and communicate over bandwidth constrained channels. For this reason, the development of efficient distributed algorit...
详细信息
This paper addresses a kernel-based learning problem for a network of agents locally observing a latent multidimensional, nonlinear phenomenon in a noisy environment. We propose a learning algorithm that requires only...
详细信息
暂无评论