Loading...
26 results
Search Results
Now showing 1 - 10 of 26
- On the Evaluation of Energy-Efficient Deep Learning Using Stacked Autoencoders on Mobile GPUsPublication . Falcao, Gabriel; Alexandre, Luís; Marques, J.; Frazão, Xavier; Maria, J.Over the last years, deep learning architectures have gained attention by winning important international detection and classification challenges. However, due to high levels of energy consumption, the need to use low-power devices at acceptable throughput performance is higher than ever. This paper tries to solve this problem by introducing energy efficient deep learning based on local training and using low-power mobile GPU parallel architectures, all conveniently supported by the same high-level description of the deep network. Also, it proposes to discover the maximum dimensions that a particular type of deep learning architecture—the stacked autoencoder—can support by finding the hardware limitations of a representative group of mobile GPUs and platforms.
- Robot Workspace Monitoring using a Blockchain-based 3D Vision ApproachPublication . Lopes, Vasco; Pereira, Nuno; Alexandre, LuísBlockchain has been used extensively for financial purposes, but this technology can also be beneficial in other contexts where multi-party cooperation, security and decentralization of the data is essential. Properties such as immutability, accessibility and non-repudiation and the existence of smart-contracts make blockchain technology very interesting in robotic contexts that require event registration or integration with Artificial Intelligence. In this paper, we propose a system that leverages blockchain as a ledger to register events and information to be processed by Oracles and uses smart-contracts to control robots by adjusting their velocity, or stopping them, if a person enters the robot working space without permission. We show how blockchain can be used in computer vision problems by interacting with multiple external parties, Oracles, that perform image analysis and how it is possible to use multiple smart-contracts for different tasks. The method proposed is shown in a scenario representing a factory environment, but since it is modular, it can be easily adapted and extended for other contexts, allowing for simple integration and maintenance.
- “Less is more”: Simplifying point clouds to improve grasping performancePublication . Lopes, Vasco; Alexandre, Luís; Fernandes, MiguelObject grasping is a task that humans do without major concerns. This results from self learning and by observing of other skilled humans doing such task with previous information. However, grasping novel objects in unknown positions for a robot is a complex task which encounters many problems, such as sub-optimal performance rates and the time consumption. In this paper we present a method that complements the state-of-the-art grasping algorithms with two segmentation steps, the first one which removes the largest planar surface in the point cloud of the world before the grasp detector receives them and the second one that complements this segmentation with another segmentation that calculates where the object is located and segments the point cloud by executing a crop around the object. The proposed method significantly improves the grasping success rate (100% improvement over the baseline approach) and simultaneously is able to reduce the time consumption by 23%.
- Image Sentiment Analysis: Experimental Evaluation of Several Deep Learning ArchitecturesPublication . Gaspar, António; Alexandre, LuísImage sentiment analysis is an important topic nowadays. It is possible to use it to classify an image at sentiment level, as negative, neutral or positive. However, to classify an image at this level is a hard challenge because its semantic meaning can represent many scenarios. In this paper, we present an analysis of several image classification methods that we evaluate to improve the state of the art in a large tweet data set.
- Distributed Learning of CNNs on Heterogeneous CPU/GPU ArchitecturesPublication . Marques, José; Falcao, Gabriel; Alexandre, LuísConvolutional Neural Networks (CNNs) have shown to be powerful classi cation tools in tasks that range from check reading to medical diagnosis, reaching close to human perception, and in some cases surpassing it. However, the problems to solve are becoming larger and more complex, which translates to larger CNNs, leading to longer training times|the computational complex part|that not even the adoption of Graphics Processing Units (GPUs) could keep up to. This problem is partially solved by using more processing units and distributed training methods that are o ered by several frameworks dedicated to neural network training, such as Ca e, Torch or TensorFlow. However, these techniques do not take full advantage of the possible parallelization o ered by CNNs and the cooperative use of heterogeneous devices with di erent processing capabilities, clock speeds, memory size, among others. This paper presents a new method for the parallel training of CNNs that can be considered as a particular instantiation of model parallelism, where only the convolutional layer is distributed. In fact, the convolutions processed during training (forward and backward propagation included) represent from 60-90% of global processing time. The paper analyzes the in uence of network size, bandwidth, batch size, number of devices, including their processing capabilities, and other parameters. Results show that this technique is capable of diminishing the training time without a ecting the classi cation performance for both CPUs and GPUs. For the CIFAR-10 dataset, using a CNN with two convolutional layers, and 500 and 1500 kernels, respectively, best speedups achieve 3:28 using four CPUs and 2:45 with three GPUs. Modern imaging datasets, larger and more complex than CIFAR-10 will certainly require more than 60-90% of processing time calculating convolutions, and speedups will tend to increase accordingly.
- Improving Grasping Performance by Segmentation of Large Planar SurfacePublication . Lopes, Vasco; Alexandre, LuísGrasping objects is a task that humans do without major concerns. This results from learning and observing other skilled humans doing such task and with previous information, unconsciously, we know how to pick up different types of objects. However, grasping novel objects in unknown positions for a robot is a complex task which encounters many problems, such as the performance rates that are not perfect and the time consumption. In this paper we present a method that complements the state-ofthe- art grasping by removing the largest planar surface of the image of the world before the grasp detector receives them. The proposed method improves the performance rate and is also capable of reducing the time consumption.
- Distribution-Based Categorization of Classifier Transfer LearningPublication . Sousa, Ricardo Gamelas; Alexandre, Luís; Santos, Jorge M.; Silva, Luís M.; Sá, Joaquim Marques deTransfer Learning (TL) aims to transfer knowledge acquired in one problem, the source problem, onto another problem, the target problem, dispensing with the bottom-up construction of the target model. Due to its relevance, TL has gained significant interest in the Machine Learning community since it paves the way to devise intelligent learning models that can easily be tailored to many different applications. As it is natural in a fast evolving area, a wide variety of TL methods, settings and nomenclature have been proposed so far. However, a wide range of works have been reporting different names for the same concepts. This concept and terminology mixture contribute however to obscure the TL field, hindering its proper consideration. In this paper we present a review of the literature on the majority of classification TL methods, and also a distribution-based categorization of TL with a common nomenclature suitable to classification problems. Under this perspective three main TL categories are presented, discussed and illustrated with examples.
- Towards AGV Optimization using ROS and Stage SimulatorPublication . Silva, Bruno Carneiro da; Alexandre, LuísAutonomous Guided Vehicles (AGV) are currently being used in industry to move materials efficiently. Simulators may be used to help calculate the right number of AGVs needed for a particular task and also which types better suit the necessity of a company. This paper analyzes the characteristics of many of the most used simulators and focus on evaluating an environment using Stage and Robot Operating System (ROS), to find experimentally if one AGV may complete a specific task taking into account eventual path blockages by random events.
- Improving SeNA-CNN by Automating Task RecognitionPublication . Zacarias, Abel; Alexandre, LuísCatastrophic forgetting arises when a neural network is not capable of preserving the past learned task when learning a new task. There are already some methods proposed to mitigate this problem in arti cial neural networks. In this paper we propose to improve upon our previous state-of-the-art method, SeNA-CNN, such as to enable the automatic recognition in test time of the task to be solved and we experimentally show that it has excellent results. The experiments show the learning of up to 4 di erent tasks with a single network, without forgetting how to solve previous learned tasks.
- Lesion Classification in Mammograms Using Convolutional Neural Networks and Transfer LearningPublication . Perre, Ana Catarina; Alexandre, Luís; Freire, Luís C.Computer-Aided Detection/Diagnosis (CAD) tools were created to assist the detection and diagnosis of early stage cancers, decreasing false negative rate and improving radiologists’ efficiency. Convolutional Neural Networks (CNNs) are one example of deep learning algorithms that proved to be successful in image classification. In this paper we aim to study the application of CNNs to the classification of lesions in mammograms. One major problem in the training of CNNs for medical applications is the large dataset of images that is often required but seldom available. To solve this problem, we use a transfer learning approach, wich is based on three different networks that were pre-trained on the Imagenet dataset. We then investigate the performance of these pre-trained CNNs and two types of image normalization to classify lesions in mammograms. The best results were obtained using the Caffe reference model for the CNN with no image normalization.
- «
- 1 (current)
- 2
- 3
- »