Data science and Artificial Intelligence

To meet our clients’ needs, we develop more efficient Artificial Intelligence (AI) engines based on cutting-edge research innovations

Artificial intelligence, more than any other sector, displays countless possibilities for use: decision making, automatic recognition, predictive engines, etc. In order to take full advantage of this potential, we identify and combine the most high-performance technologies for you: deep learning, machine learning for image and natural language processing, Bayesian networks, etc.

Even if your data is unstructured and limited in quantity, albeit in text, image or graph form, our multidisciplinary team will assist you in coming up with the best strategy for making use of it. We are there to guide you throughout the entire project, from the data audit up to the designing of turnkey AI solutions that include custom-made algorithms.

Data engineering

While using data is always good, exploiting relevant, high-quality data is infinitely better. Since it is often unstructured, the available data is not organised in formats that can be interpreted by machines. So how can you use it in your AI projects? Our specialists make sure that your data, no matter the type or quantity, gets used according to your objectives.

In a nutshell
Data collectionData processingData mining

Data capture

  • Data collection: images, texts, machine data, structured or unstructured data
  • Data acquisition campaign by sensors, video acquisition…
  • Decision support for data localisation and storage: edge / fogg / cloud computing

Pre-processing of data

  • Diagnosis of data quality / quantity
  • Data augmentation methods
  • Refining and preparation of data: removing noise, any information not useful for the algorithm
  • Structuring of data, even heterogeneous data
  • Manual annotation of data to allow its exploitation
  • Anonymisation of data

Data mining

  • Development of data mining tools
  • Extraction of relevant information from a large dataset
  • Automatic or semi-automatic data retrieval methods

Deep learning – Computer vision

Our deep learning algorithms are specifically designed in order to allow you to use your sensor and camera data. Destined for automatic image recognition, robot vision, language processing and predictive maintenance, they allow our clients to achieve sustainable innovation.

In a nutshell
Computer visionMachine learningVISUAL PERCEPTIONCOGNITIVE ALGORITHMSPERCEPTION MODELLINGSENSORY EXPERIENCE

Deep-learning

  • Quantitative and qualitative audit of data
  • Data augmentation to train artificial neural networks (ANN)
  • Optimisation of data pre-processing steps
  • Algorithmic optimisation

Computer vision for robotics

  • Choice of cameras, sensors, detection systems
  • Tests and metrological evaluation of field sensing tools
  • Algorithmic development

Image recognition

  • Visual data: images, videos, 2D and 3D.
  • Detection and recognition of shapes and objects in images
  • Recognition of people and activities
  • Classification, segmentation of images
  • Comprehension: automatic analysis of scenes
  • Decision-making (e.g. automating the control of an industrial robot)

Areas of application

  • Autonomous driving – semi-automatic driving for agriculture
  • Land and underwater inspection
  • Machine vision – defect detection
  • Medicine – medical image analysis
  • Defence and security

Predictive analytics – probabilistic modelling

Probabilistic graphical models represent your complex systems based on the knowledge of your industry experts and a reduced data set. Our tailor-made models enable anomaly detection, predictive maintenance, decision-making support, etc. This AI, possessing a comprehensive logic, improves over time with increasing feedback.

In a nutshell
Process miningBayesian networkprobabilistic graphical model

Prediction – Decision Support

  • Prediction of indicators from observed data
  • Assessment of a risk incurred, prediction of failures…
  • Detection of aberrant behaviour in real time and sending of alert messages

Probabilistic modelling

  • Modelling from flat, structured or partially structured data
  • Construction from data, business rules, expert knowledge
  • Bayesian networks
  • Process mining

Explainable reasoning

  • Responsible AI (white-box) : models with understandable behaviors which
  • Modeling based on human and business experts’ experience
  • Probabilistic graphical models

Small data – data engineering

  • Development of an initial model based on the knowledge of business experts, with no other data available
  • Prediction models can be continuously improved through feedback
  • Support for data exploitation: audit, refinement, preparation, structuring
  • Data mining

Web interfaces, applications & software

  • Inclusion of our Pilgrim® AI engine in your business applications
  • Development of custom web/mobile applications integrating our AI engine

Predictive maintenance

  • Definition of critical parameters
  • Data pre-processing, qualitative and quantitative data analysis
  • Machine learning, statistical modelling, probabilistic modelling
  • Data visualisation
  • Real-time monitoring and alerts
  • Embedded software and web and mobile application development

Natural language processing

Our experts are there to guide you in making the most of your text-based and general language-based data: conversations, written correspondence, etc. We provide you with automatic text mining solutions, including the automated search and analysis of sentiments and opinions.

In a nutshell
SEMANTIC ANALYSISOpinion miningText miningDeep learningMachine learning

Semantic analysis

  • Text mining
  • Opinion and sentiment analysis
  • Nomenclature construction, domain terminology
  • Multilingual document processing

Data science – Algorithm

  • Audit of available data: typology, recalibration of texts to be analysed
  • Data exploitation and pre-processing strategy: annotation of texts, constitution of a corpus
  • Alignment and automated processing of heterogeneous data: handwriting, spoken data, equations, images, etc.
  • Modelling and development of algorithms adapted to the data and the client’s needs
  • Testing and analysis of the quality of the algorithm produced, on a data set

Perception testing and video encoding strategy

The world leaders in broadcasting call upon our experts in order to assist them in optimising image compression without impairing the quality of the user experience on their platforms. By combining bio-inspired algorithms and perception tests using sensors, they objectively analyse the performance of a fixed visual representation (image) or a moving one (video).

In a nutshell
TESTS & EXPERIMENTSALGORITHMS AND MODELLINGIMAGE ENCODING

Tests & experiments

  • Development of experimental evaluation protocols and new setting methods
  • Tests of sensory perception: in the laboratory, on site, in immersion or on a crowdsourcing platform
  • Psycho-physiological measurements: oculometry, EEG neural sensors

Algorithms and modelling

  • Collection and analysis of data from psycho-physiological measurements
  • Design of models for the objectification of user perception
  • Development of visual attention models: salience model and prediction of object of perceptual interest
  • Verification of prediction models of perception

Image encoding strategies

  • Encoding methods
  • Development of compression algorithms
  • Advice to development teams

Application development

Our team of developers design the application layer needed to make your projects come to life. We integrate the custom-made AI engines designed by our data scientists and combine them with other technologies (augmented reality, indoor geolocation, etc.) where necessary, in order to offer you high value-added solutions that are perfectly suited to your needs

In a nutshell
Proof of conceptAI integrationWeb & mobile appSoftware – HMIInnovative interfaces

POC (Proof of Concept)

  • Development of a functional tool to prove the concept of your project
  • Tests on a dataset to validate the efficiency of the artificial intelligence according to your needs

Integration of AI models

  • Integration at the server level (node.js)
  • Creation of APIs (Application Programming Interface) to enable the integration of your solution with your information system (IS)
  • Development of SaaS solutions on secure cloud platforms

UX design

  • A design approach focused on the use and habits of your customers
  • Graphic and web design, in respect of the brand identity

Multi-media software solutions

  • Creation of iOS and Android mobile applications, web platforms
  • Web back-office to manage your resources and results
  • UX design: attention to the ergonomics of applications
  • Native applications

Measurement, interfacing and visualisation tools

  • Design of unique experiences combining new technologies
  • Physiological sensors (neural helmets)
  • Augmented reality
  • Indoor geolocation (beacons)
  • Bluetooth technologies
  • Visual and voice recognition devices (e.g. Kinect, voice assistants…)

Augmented reality

  • Development of augmented reality interfaces using arKit, ARCore, Vuforia
  • Realistic visualisation of a product, an image, a video, explanatory texts in your environments

Software – HMI

  • Custom software and HMI development for automation and robotics
  • Specific developments for test benches and applications for measurement and automation
  • Mathematical development and calculation algorithms for the automated control of robots and machines

Engineering at CAPACITÉS

Our Data Science and AI experts perform their engineering duties in the very research laboratories of the University of Nantes themselves, most notably at LS2N. They strive to combine technical resources and scientific discovery to meet your R&D needs.

Challenge we’ve met

Encoding algorithms: Netflix chooses Capacités and LS2N

The Californian SVOD giant noticed the research that Capacités and the LS2N had conducted in AI and visual perception modeling.

Read this success story

Facebook Chooses Capacités to Optimize Video Encoding

Facebook chose experts in visual perception at Capacités and the LS2N laboratory in Nantes to adapt its automatic encoding process to this new video content.

Read this success story

This site uses cookies and gives you control over what you want to enable