Sébastien Lafond

Muhammad Junaid Haris, Aanchal Upreti, Melih Kurtaran et al.

The problem of gender bias is highly prevalent and well known. In this paper, we have analysed the portrayal of gender roles in English movies, a medium that effectively influences society in shaping people's beliefs and opinions. First, we gathered scripts of films from different genres and derived sentiments and emotions using natural language processing techniques. Afterwards, we converted the scripts into embeddings, i.e. a way of representing text in the form of vectors. With a thorough investigation, we found specific patterns in male and female characters' personality traits in movies that align with societal stereotypes. Furthermore, we used mathematical and machine learning techniques and found some biases wherein men are shown to be more dominant and envious than women, whereas women have more joyful roles in movies. In our work, we introduce, to the best of our knowledge, a novel technique to convert dialogues into an array of emotions by combining it with Plutchik's wheel of emotions. Our study aims to encourage reflections on gender equality in the domain of film and facilitate other researchers in analysing movies automatically instead of using manual approaches.

Minahil Raza, Hanna Prokopova, Samir Huseynzade et al.

Obstacle detection is a fundamental capability of an autonomous maritime surface vessel (AMSV). State-of-the-art obstacle detection algorithms are based on convolutional neural networks (CNNs). While CNNs provide higher detection accuracy and fast detection speed, they require enormous amounts of data for their training. In particular, the availability of domain-specific datasets is a challenge for obstacle detection. The difficulty in conducting onsite experiments limits the collection of maritime datasets. Owing to the logistic cost of conducting on-site operations, simulation tools provide a safe and cost-efficient alternative for data collection. In this work, we introduce SimuShips, a publicly available simulation-based dataset for maritime environments. Our dataset consists of 9471 high-resolution (1920x1080) images which include a wide range of obstacle types, atmospheric and illumination conditions along with occlusion, scale and visible proportion variations. We provide annotations in the form of bounding boxes. In addition, we conduct experiments with YOLOv5 to test the viability of simulation data. Our experiments indicate that the combination of real and simulated images improves the recall for all classes by 2.9%.

Bashir Alam, Masa Cirkovic, Mete Harun Akcay et al.

Imbalanced datasets pose a considerable challenge in training deep learning (DL) models for medical diagnostics, particularly for segmentation tasks. Imbalance may be associated with annotation quality limited annotated datasets, rare cases, or small-scale regions of interest (ROIs). These conditions adversely affect model training and performance, leading to segmentation boundaries which deviate from the true ROIs. Traditional loss functions, such as Binary Cross Entropy, replicate annotation biases and limit model generalization. We propose a novel, statistically driven, conditionally adaptive loss function (CALF) tailored to accommodate the conditions of imbalanced datasets in DL training. It employs a data-driven methodology by estimating imbalance severity using statistical methods of skewness and kurtosis, then applies an appropriate transformation to balance the training dataset while preserving data heterogeneity. This transformative approach integrates a multifaceted process, encompassing preprocessing, dataset filtering, and dynamic loss selection to achieve optimal outcomes. We benchmark our method against conventional loss functions using qualitative and quantitative evaluations. Experiments using large-scale open-source datasets (i.e., UPENN-GBM, UCSF, LGG, and BraTS) validate our approach, demonstrating substantial segmentation improvements. Code availability: https://anonymous.4open.science/r/MICCAI-Submission-43F9/.

Joel Öhrling, Sébastien Lafond, Dragos Truscan

In this paper we evaluate the use of system identification methods to build a thermal prediction model of heterogeneous SoC platforms that can be used to quickly predict the temperature of different configurations without the need of hardware. Specifically, we focus on modeling approaches that can predict the temperature based on the clock frequency and the utilization percentage of each core. We investigate three methods with respect to their prediction accuracy: a linear state-space identification approach using polynomial regressors, a NARX neural network approach and a recurrent neural network approach configured in an FIR model structure. We evaluate the methods on an Odroid-XU4 board featuring an Exynos 5422 SoC. The results show that the model based on polynomial regressors significantly outperformed the other two models when trained with 1 hour and 6 hours of data.

Joel Öhrling, Dragos Truscan, Sebastien Lafond

The management of the energy consumption and thermal dissipation of multi-core heterogeneous platforms is becoming increasingly important as it can have direct impact on the platform performance. This paper discusses an approach that enables fast exploration and validation of heterogeneous system on chips (SoCs) platform configurations with respect to their thermal dissipation. Such platforms can be configured to find the optimal trade-off between performance and power consumption. This directly reflects in the head dissipation of the platform, which when increases over a given threshold will actually decrease the performance of the platform. Therefore, it is important to be able to quickly probe and explore different configurations and identify the most suitable one. However, this task is hindered by the large space of possible configurations of such platforms and by the time required to benchmark each configurations. As such, we propose an approach in which we construct a model of the thermal dissipation of a given platform using a system identification methods and then we use this model to explore and validate different configurations. The approach allows us to decrease the exploration time with several orders of magnitude. We exemplify the approach on an Odroid-XU4 board featuring an Exynos 5422 SoC.

Ivan Porres, Hergys Rexha, Sébastien Lafond

In this paper we present a novel algorithm for automatic performance testing that uses an online variant of the Generative Adversarial Network (GAN) to optimize the test generation process. The objective of the proposed approach is to generate, for a given test budget, a test suite containing a high number of tests revealing performance defects. This is achieved using a GAN to generate the tests and predict their outcome. This GAN is trained online while generating and executing the tests. The proposed approach does not require a prior training set or model of the system under test. We provide an initial evaluation the algorithm using an example test system, and compare the obtained results with other possible approaches. We consider that the presented algorithm serves as a proof of concept and we hope that it can spark a research discussion on the application of GANs to test generation.

Hergys Rexha, Sebastien Lafond

As data being produced by IoT applications continues to explode, there is a growing need to bring computing power closer to the source of the data to meet the response time, power dissipation and cost goals of performance-critical applications in various domains like the Industrial Internet of Things (IIoT), Automated Driving, Medical Imaging or Surveillance among others. This paper proposes a data collection and utilization framework that allows runtime platform and application data to be sent to an edge and cloud system via data collection agents running close to the platform. Agents are connected to a cloud system able to train AI models to improve overall energy efficiency of an AI application executed on an edge platform. In the implementation part, we show the benefits of FPGA-based platform for the task of object detection. Furthermore, we show that it is feasible to collect relevant data from an FPGA platform, transmit the data to a cloud system for processing and receiving feedback actions to execute an edge AI application energy efficiently. As future work, we foresee the possibility to train, deploy and continuously improve a base model able to efficiently adapt the execution of edge applications.

Namid Stillman, Igor Balaz, Antisthenis Tsompanas et al.

We present the EVONANO platform for the evolution of nanomedicines with application to anti-cancer treatments. EVONANO includes a simulator to grow tumours, extract representative scenarios, and then simulate nanoparticle transport through these scenarios to predict nanoparticle distribution. The nanoparticle designs are optimised using machine learning to efficiently find the most effective anti-cancer treatments. We demonstrate our platform with two examples optimising the properties of nanoparticles and treatment to selectively kill cancer cells over a range of tumour environments.

Ivan Porres, Sepinoud Azimi, Sébastien Lafond et al.

This paper explores the state of the art on to methods to verify and validate navigation algorithms for autonomous surface ships. We perform a systematic mapping study to find research works published in the last 10 years proposing new algorithms for autonomous navigation and collision avoidance and we have extracted what verification and validation approaches have been applied on these algorithms. We observe that most research works use simulations to validate their algorithms. However, these simulations often involve just a few scenarios designed manually. This raises the question if the algorithms have been validated properly. To remedy this, we propose the use of a systematic scenario-based testing approach to validate navigation algorithms extensively.