Despite the effectiveness of vision-language supervised fine-tuning in enhancing the performance of vision large language models(VLLMs), existing visual instruction tuning datasets include the following limitations.(1...
详细信息
Despite the effectiveness of vision-language supervised fine-tuning in enhancing the performance of vision large language models(VLLMs), existing visual instruction tuning datasets include the following limitations.(1) Instruction annotation quality: despite existing VLLMs exhibiting strong performance,instructions generated by those advanced VLLMs may still suffer from inaccuracies, such as hallucinations.(2) Instructions and image diversity: the limited range of instruction types and the lack of diversity in image data may impact the model's ability to generate diversified and closer to real-world scenarios outputs. To address these challenges, we construct a high-quality, diverse visual instruction tuning dataset MMInstruct,which consists of 973k instructions from 24 domains. There are four instruction types: judgment, multiplechoice, long visual question answering, and short visual question answering. To construct MMInstruct, we propose an instruction generation data engine that leverages GPT-4V, GPT-3.5, and manual correction. Our instruction generation engine enables semi-automatic, low-cost, and multi-domain instruction generation at 1/6 the cost of manual construction. Through extensive experiment validation and ablation experiments,we demonstrate that MMInstruct could significantly improve the performance of VLLMs, e.g., the model fine-tuning on MMInstruct achieves new state-of-the-art performance on 10 out of 12 benchmarks. The code and data shall be available at https://***/yuecao0119/MMInstruct.
Trajectory contains spatial-data generated from traces of moving objects like people, animals, etc. Community generated from trajectories portrays common behaviour. Trajectory clustering based on community-detection i...
详细信息
Agriculture encompasses a way of life and a profession for the general population. Most global traditions and cultures revolve around agriculture. With the help of advanced farming, agriculture may become more profita...
详细信息
In this paper, the computation of graph Fourier transform centrality (GFTC) of complex network using graph filter is presented. For conventional computation method, it needs to use the non-sparse transform matrix of g...
详细信息
In batch production systems, detecting low-yield machines is essential for minimizing the production of defective pieces, which is a complex problem that currently requires multiple experts, considerable capital, or a...
详细信息
A new online scheduling algorithm is proposed for photovoltaic(PV)systems with battery-assisted energy storage systems(BESS).The stochastic nature of renewable energy sources necessitates the employment of BESS to bal...
详细信息
A new online scheduling algorithm is proposed for photovoltaic(PV)systems with battery-assisted energy storage systems(BESS).The stochastic nature of renewable energy sources necessitates the employment of BESS to balance energy supplies and demands under uncertain weather *** proposed online scheduling algorithm aims at minimizing the overall energy cost by performing actions such as load shifting and peak shaving through carefully scheduled BESS charging/discharging *** scheduling algorithm is developed by using deep deterministic policy gradient(DDPG),a deep reinforcement learning(DRL)algorithm that can deal with continuous state and action *** of the main contributions of this work is a new DDPG reward function,which is designed based on the unique behaviors of energy *** new reward function can guide the scheduler to learn the appropriate behaviors of load shifting and peak shaving through a balanced process of exploration and *** new scheduling algorithm is tested through case studies using real world data,and the results indicate that it outperforms existing algorithms such as Deep *** online algorithm can efficiently learn the behaviors of optimum non-casual off-line algorithms.
The increasing dependence on smartphones with advanced sensors has highlighted the imperative of precise transportation mode classification, pivotal for domains like health monitoring and urban planning. This research...
详细信息
The increasing dependence on smartphones with advanced sensors has highlighted the imperative of precise transportation mode classification, pivotal for domains like health monitoring and urban planning. This research is motivated by the pressing demand to enhance transportation mode classification, leveraging the potential of smartphone sensors, notably the accelerometer, magnetometer, and gyroscope. In response to this challenge, we present a novel automated classification model rooted in deep reinforcement learning. Our model stands out for its innovative approach of harnessing enhanced features through artificial neural networks (ANNs) and visualizing the classification task as a structured series of decision-making events. Our model adopts an improved differential evolution (DE) algorithm for initializing weights, coupled with a specialized agent-environment relationship. Every correct classification earns the agent a reward, with additional emphasis on the accurate categorization of less frequent modes through a distinct reward strategy. The Upper Confidence Bound (UCB) technique is used for action selection, promoting deep-seated knowledge, and minimizing reliance on chance. A notable innovation in our work is the introduction of a cluster-centric mutation operation within the DE algorithm. This operation strategically identifies optimal clusters in the current DE population and forges potential solutions using a pioneering update mechanism. When assessed on the extensive HTC dataset, which includes 8311 hours of data gathered from 224 participants over two years. Noteworthy results spotlight an accuracy of 0.88±0.03 and an F-measure of 0.87±0.02, underscoring the efficacy of our approach for large-scale transportation mode classification tasks. This work introduces an innovative strategy in the realm of transportation mode classification, emphasizing both precision and reliability, addressing the pressing need for enhanced classification mechanisms in an eve
Delay tolerant wireless sensor networks(DTWSN)is a class of wireless network that finds its deployment in those application scenarios which demand for high packet delivery ratio while maintaining minimal overhead in o...
详细信息
Delay tolerant wireless sensor networks(DTWSN)is a class of wireless network that finds its deployment in those application scenarios which demand for high packet delivery ratio while maintaining minimal overhead in order to prolong network lifetime;owing to resource-constrained nature of *** fundamental requirement of any network is routing a packet from its source to *** of a routing algorithm depends on the number of network parameters utilized by that routing *** the recent years,various routing protocol has been developed for the delay tolerant networks(DTN).A routing protocol known as spray and wait(SnW)is one of the most widely used routing algorithms for *** this paper,we study the SnW routing protocol and propose a modified version of it referred to as Pentago SnW which is based on pentagonal number *** to binary SnW shows promising results through simulation using real-life scenarios of cars and pedestrians randomly moving on a map.
This article proposes a proactive crowdsourced monitoring and sensing (PCMS) framework with the designed Smart iBeacon device to accurately recognize the activities of an equipped target, exclusively customize the rec...
详细信息
The security of the wireless sensor network-Internet of Things(WSN-IoT)network is more challenging due to its randomness and self-organized *** detection is one of the key methodologies utilized to ensure the security...
详细信息
The security of the wireless sensor network-Internet of Things(WSN-IoT)network is more challenging due to its randomness and self-organized *** detection is one of the key methodologies utilized to ensure the security of the *** intrusion detection mechanisms have issues such as higher misclassification rates,increased model complexity,insignificant feature extraction,increased training time,increased run time complexity,computation overhead,failure to identify new attacks,increased energy consumption,and a variety of other factors that limit the performance of the intrusion system *** this research a security framework for WSN-IoT,through a deep learning technique is introduced using Modified Fuzzy-Adaptive DenseNet(MF_AdaDenseNet)and is benchmarked with datasets like NSL-KDD,UNSWNB15,CIDDS-001,Edge IIoT,Bot *** this,the optimal feature selection using Capturing Dingo Optimization(CDO)is devised to acquire relevant features by removing redundant *** proposed MF_AdaDenseNet intrusion detection model offers significant benefits by utilizing optimal feature selection with the CDO *** results in enhanced Detection Capacity with minimal computation complexity,as well as a reduction in False Alarm Rate(FAR)due to the consideration of classification error in the fitness *** a result,the combined CDO-based feature selection and MF_AdaDenseNet intrusion detection mechanism outperform other state-of-the-art techniques,achieving maximal Detection Capacity,precision,recall,and F-Measure of 99.46%,99.54%,99.91%,and 99.68%,respectively,along with minimal FAR and Mean Absolute Error(MAE)of 0.9%and 0.11.
暂无评论