A5039374801- Advanced Neural Network Applications
- Anomaly Detection Techniques and Applications
- Video Analysis and Summarization
- Blockchain Technology Applications and Security
- Domain Adaptation and Few-Shot Learning
- Robotics and Sensor-Based Localization
- Advanced Text Analysis Techniques
- IoT and Edge/Fog Computing
- Privacy-Preserving Technologies in Data
- Music and Audio Processing
- Mobile Crowdsensing and Crowdsourcing
- Reinforcement Learning in Robotics
- Speech and dialogue systems
- Olfactory and Sensory Function Studies
- CCD and CMOS Imaging Sensors
- Adversarial Robustness in Machine Learning
- Big Data and Business Intelligence
- Speech Recognition and Synthesis
- International Relations in Latin America
- Advanced Memory and Neural Computing
- Advanced Image and Video Retrieval Techniques
- Polydiacetylene-based materials and applications
- Handwritten Text Recognition Techniques
- Machine Learning and Data Classification
- Robotic Path Planning Algorithms
VERSES (United States)
2024
ME Association
2022-2023
The Geneva Association
2023
ETH Zurich
2018-2023
HES-SO Arc
2019-2021
University of Bologna
2021
Standard-sized autonomous vehicles have rapidly improved thanks to the breakthroughs of deep learning. However, scaling driving mini-vehicles poses several challenges due their limited on-board storage and computing capabilities. Moreover, systems lack robustness when deployed in dynamic environments where underlying distribution is different from learned during training. To address these challenges, we propose a closed-loop learning flow for that includes target deployment environment...
Next generation of embedded Information and Communication Technology (ICT) systems are interconnected collaborative able to perform autonomous tasks. The remarkable expansion the ICT market, together with rise breakthroughs Artificial Intelligence (AI), have put focus on Edge as it stands one keys for next technological revolution: seamless integration AI in our daily life. However, training deployment custom solutions devices require a fine-grained data, algorithms, tools achieve high...
The spread of deep learning on embedded devices has prompted the development numerous methods to optimise deployment neural networks (DNN). Works have mainly focused on: i) efficient DNN architectures, ii) network optimisation techniques such as pruning and quantisation, iii) optimised algorithms speed up execution most computational intensive layers and, iv) dedicated hardware accelerate data flow computation. However, there is a lack research cross-level space approaches becomes too large...
Deep Learning is moving to edge devices, ushering in a new age of distributed Artificial Intelligence (AI). The high demand computational resources required by deep neural networks may be alleviated approximate computing techniques, and most notably reduced-precision arithmetic with coarsely quantized numerical representations. In this context, Bonseyes comes as an initiative enable stakeholders bring AI low-power autonomous environments such as: Automotive, Medical Healthcare Consumer...
Abstract Sarcomas are mesenchymal cancers which often show an aggressive behavior and patient survival largely depends on early detection. In last years, much attention has been given to the fact that cancer patients release specific odorous volatile organic compounds (VOCs) can be efficiently detected by properly trained sniffer dogs. Here, we have evaluated for first time ability of dogs (n = 2) detect osteosarcoma cell cultures samples. One two was successfully discriminate...
Deep Learning is increasingly being adopted by industry for computer vision applications running on embedded devices. While Convolutional Neural Networks' accuracy has achieved a mature and remarkable state, inference latency throughput are major concern especially when targeting low-cost low-power platforms. CNNs' may become bottleneck adoption industry, as it crucial specification many real-time processes. Furthermore, deployment of CNNs across heterogeneous platforms presents...
The size and diversity of the training datasets directly influences decision-making process AI models. Therefore, there is an immense need for massive diverse to enhance deployment applications. Crowdsourcing marketplaces provide a fast reliable alternative laborious data collection process. However, existing crowdsourcing are either centralized or do not fully sovereignty. By contrast, this work proposes decentralized platform through prototypical implementation along with active...
The accuracy of a keyword spotting model deployed on embedded devices often degrades when the system is exposed to real environments with significant noise. In this paper, we explore methodology for tailoring on-site noises through on-device domain adaptation, while accounting edge computing-associated costs. We show that improvements by up 18 % can be obtained specialising difficult, previously unseen noise types, power budget in Watt range, storage requirement 1.1 GB. also demonstrate an...
Today's rapidly changing marketplaces are constantly bringing new ways of transforming business operations and require companies to be flexible dynamic. Toward this direction the use Artificial Intelligence (AI) tools techniques has potential bring significant value by increasing their efficiency. Companies already benefiting from AI solutions such as prescriptive analytics enhance customers" experience or achieve optimal limited resources. This paper analyses opportunities challenges...
Bonseyes is an Artificial Intelligence (AI) platform composed of a Data Marketplace, Deep Learning Toolbox, and Developer Reference Platforms with the aim facilitating tech non-tech companies rapid adoption AI as enabler for their business. provides methods tools to speed up development deployment solutions on low power Internet Things (IoT) devices, embedded computing systems, data centre servers. In this work, we address integration applications in wider enterprise application landscape...
Standard-size autonomous navigation vehicles have rapidly improved thanks to the breakthroughs of deep learning. However, scaling driving low-power systems deployed on dynamic environments poses several challenges that prevent their adoption. To address them, we propose a closed- loop learning flow for mini-vehicles includes target environment in-the-loop. We leverage family compact and high-throughput tinyCNNs control mini- vehicle, which learn in by imitating computer vision algorithm,...
Visual inspection plays a pivotal role in numerous industrial production processes, and the pursuit of automation has surged with rise deep learning convolutional neural networks (CNNs). Therein, deployment visual CNNs on resource-constrained edge devices stands as critical problem these are most affordable well-suited for many applications, e.g., chains. Nonetheless, it faces challenges meeting computational demands CNN models. Consequently, optimizing models efficient operation such...
Keyword spotting accuracy degrades when neural networks are exposed to noisy environments. On-site adaptation previously unseen noise is crucial recovering loss, and on-device learning required ensure that the process happens entirely on edge device. In this work, we propose a fully domain system achieving up 14% gains over already-robust keyword models. We enable with less than 10 kB of memory, using only 100 labeled utterances recover 5% after adapting complex speech noise. demonstrate can...
Predicting future trajectories of nearby objects, especially under occlusion, is a crucial task in autonomous driving and safe robot navigation. Prior works typically neglect to maintain uncertainty about occluded objects only predict observed using high-capacity models such as Transformers trained on large datasets. While these approaches are effective standard scenarios, they can struggle generalize the long-tail, safety-critical scenarios. In this work, we explore conceptual framework...
The demand for large-scale diverse datasets is rapidly increasing due to the advancements in AI services impacting day-to-day life. However, gathering such massive still remains a critical challenge service engineering pipeline, especially computer vision domain where labeled data scarce. Rather than isolated collection, crowdsourcing techniques have shown promising potential achieve collection task time and cost-efficient manner. In existing marketplaces, crowd works fulfill...

 Artificial Intelligence (AI) is becoming increasingly important and pervasive in the modern world. The widespread adoption of AI algorithms reflected extensive range HW devices on which they can be deployed, from high-performance computing nodes to low-power embedded devices. Given large set heterogeneous resources where finding most suitable device its con- figuration challenging, even for experts.
 We propose a data-driven approach assist adopters developers choosing optimal...