In this work, we present a multi-modal model for commercial product classification, that combines features extracted by multiple neural network models from textual (Camem-BERT and FlauBERT) and visual data (SE-ResNeXt...
详细信息
A popular track of network compression approach is Quantization aware Training (QAT), which accelerates the forward pass during the neural network training and inference. However, not much prior efforts have been made...
A popular track of network compression approach is Quantization aware Training (QAT), which accelerates the forward pass during the neural network training and inference. However, not much prior efforts have been made to quantize and accelerate the backward pass during training, even though that contributes around half of the training time. This can be partly attributed to the fact that errors of low-precision gradients during backward cannot be amortized by the training objective as in the QAT setting. In this work, we propose to solve this problem by incorporating the gradients into the computation graph of the next training iteration via a hypernetwork. Various experiments on CIFAR-10 dataset with different CNN network architectures demonstrate that our hypernetwork-based approach can effectively reduce the negative effect of gradient quantization noise and successfully quantizes the gradients to INT4 with only 0.64 accuracy drop for VGG-16 on CIFAR-10.
This paper presents a multi-objective version of the Cat Swarm Optimization Algorithm called the Grid-based Multiobjective Cat Swarm Optimization Algorithm (GMOCSO). Convergence and diversity preservation are the two ...
详细信息
Elliptic curve cryptosystem is a widely-used public key cryptosystem that is vulnerable to physical attacks, especially to Side Channel Analysis (SCA) attacks. This paper investigates a version of the atomic patterns ...
Elliptic curve cryptosystem is a widely-used public key cryptosystem that is vulnerable to physical attacks, especially to Side Channel Analysis (SCA) attacks. This paper investigates a version of the atomic patterns $\boldsymbol{kP}$ algorithm for elliptic curves over prime finite fields. The atomic patterns for Elliptic Curve (EC) point doublings and additions are well-known countermeasures against simple SCA attacks. We implemented a binary left-to-right $\boldsymbol{kP}$ algorithm corresponding to the atomic patterns as proposed by Rondepierre and using the FLECC open-source cryptographic library. The library provides constant-time functions for the implementation of different mathematic operations in prime finite fields, making it suitable for SCA-aware implementations of cryptographic protocols, at a theoretical level. We were able to reveal the scalar k processed during a $\boldsymbol{kP}$ execution analyzing the measured electromagnetic trace of the $\boldsymbol{kP}$ execution. We determined at least two parts in the measured trace where the distinction between EC point doublings and EC point additions is good observable.
The Network Data Analytics Function (NWDAF) within the 5G core is not an inherent feature of open-source 5G cores, making its implementation necessary based on provider demands. However, for those seeking to integrate...
详细信息
ISBN:
(数字)9798350376562
ISBN:
(纸本)9798350376579
The Network Data Analytics Function (NWDAF) within the 5G core is not an inherent feature of open-source 5G cores, making its implementation necessary based on provider demands. However, for those seeking to integrate NWDAF into the core but lacking expertise in core architecture or application programming interface (API) integration, this task can be daunting, potentially resulting in the abandonment of data analytics implementation efforts. While the current version of NWDAF specifies certain use cases, it does not provide alternatives for developing new analytics scenarios. To fill this gap, this paper proposes a framework for 5G network data analytics designed to facilitate algorithmic modifications and the addition of new analytical contexts. The framework was developed considering the 3GPP technical specifications related to data analytics. Its evaluation was performed through a use case focused on cybersecurity anomaly detection applied to a 5G network core. The results demonstrated that the framework simplifies installation, increases flexibility for new algorithms, and integrates seamlessly into the 5G network core, effectively addressing previously identified shortcomings.
Variational quantum algorithms, inspired by neural networks, have become a novel approach in quantum computing. However, designing efficient parameterized quantum circuits remains a challenge. Quantum architecture sea...
详细信息
Equipping the society with an evacuation route assistant is important as a preventive action in order to minimize anxiety as well as number of victims when disaster strike. Therefore, research on best route determinat...
Equipping the society with an evacuation route assistant is important as a preventive action in order to minimize anxiety as well as number of victims when disaster strike. Therefore, research on best route determination to the shelter is a noble task. Evacuation routes mean routes to pass in order to reach the gathering point or shelter safely. This study aims to implement the particle swarm optimization (PSO) algorithm in finding the optimal path to the designated shelter. PSO is a metaheuristic algorithm inspired by the behaviour of flocks of birds or fish to optimize the search for the best route. Taking into account the details of the PSO implementation, including initialization of the random particle position and velocity, selection of the best particle (Pbest) and Global Best (Gbest), and updating the speed and position of the particle using the specified formula. The objective function is recalculated based on the total distance of the particle route as well as the condition weight of the affected areas. The results showed that out of ten experiments with 100 iterations, six valid routes were found with the highest fitness value of 1,935, which indicates that PSO can provide optimal solutions in the context of finding flood evacuation routes.
Playing games can make people feel fun. Players can be more fun if you have a lot of variations, for example, variations of items, maps, or enemies. Map variations can be fulfilled by implementing a dynamic labyrinth ...
详细信息
The dynamic and uneven terrain of our environment introduces complexity for locomotion vehicles to navigate easily across different settings. Wheelchairs belong to the category of vehicles that require locomotion acro...
详细信息
ISBN:
(数字)9798331511241
ISBN:
(纸本)9798331511258
The dynamic and uneven terrain of our environment introduces complexity for locomotion vehicles to navigate easily across different settings. Wheelchairs belong to the category of vehicles that require locomotion across various uneven terrains, such as stairs and other partitions. Therefore, the aim of this work is to design a wheel that can travel through various environments without using complex mechanisms. This aim was achieved by creating a novel transformable wheel. This wheel can turn into a legged wheel without using a complex mechanism and, when unnecessary, transform back into a normal wheel. Several simulations were carried out to understand the design's capability to overcome obstacles. In this case, the stair was the obstacle, and the results illustrated that the wheel is capable of climbing stairs without serious issues. Different parameters, such as speed and force, were examined with various values. The results showed that the weight on the wheelchair plays a significant role in climbing. For instance, with lighter weight, the structure moves without major slipping. The roughness of the surface is also related to the wheelchair's ability to reach the top of the stairs. A finer depth of cuts on the surface of the stairs increases the chance of the legged wheel firmly gripping and moving up the stairs.
Image retargeting involves the adjustment of an image's dimensions to ensure that its content and visual quality are preserved when the image is resized to fit various screens or devices. This process retains all ...
Image retargeting involves the adjustment of an image's dimensions to ensure that its content and visual quality are preserved when the image is resized to fit various screens or devices. This process retains all essential visual elements and details. Different techniques have been developed for this purpose, including cropping (CR), scaling (SCL), seam carving (SC), warping (WARP), scale-and-stretch (SNS), multi-operator (MULTI), and shift-map (SM). However, determining the most suitable method for retargeting a specific image with particular dimensions remains a challenge. Therefore, this research introduces initial work on developing CNN based deep learning model and a transfer learning model based on InceptionV3,to predict the optimal retargeting method for a given image and resolution. The study employed a dataset consisting of 46,716 images with varying resolutions, created using different retargeting techniques, categorized into six groups. Results demonstrates a promising effectiveness of the proposed approach for selecting the appropriate retargeting techniques.
暂无评论