- Sectors
- Aerospace & Defense
- Big science
- Biotechnology
- Fintech
- Work at ARQUIMEA
- Insights
Investigating how deep learning approach to Artificial Intelligence (AI) can help streamlining and automating 3D scenes and objects scanning and model building. The final goal is to get a 3D model format that is much lighter, more cost-effective and more photorealistic than conventional ones–based on point clouds, meshes or voxels–through the so-called “neural fields.”
Artificial Intelligence arises from the use of computer mechanisms that aim to replicate the biological behavior of the brain by adapting and learning from large amounts of data. The structures that carry out these processes are called Artificial Neural Networks (ANN) and are composed of a series of layers that include artificial neurons. These receive data from external sources, process them conveniently and, finally, transfer the information to one or more neurons in the network. In this way, an ANN can be trained by means of large data sets and, from there, acquire the ability to infer conclusions through input data that are a priori unknown.
Photonic technologies, despite the outstanding capabilities and potential they have demonstrated for Artificial Intelligence, is an area of research, commonly known as Neuromorphic Photonics, that is still in a nascent and long development stage.
Our projects advance in the design, development and validation of a new neural network architecture that can eventually be exploited in Artificial Intelligence applications.
The main focus will be on the optimization of basic functionalities, as is the case of the weighting process, and on the proposal of novel architecture schemes that offer a clear competitive advantage in terms of speed, consumption, capacity and scalability in relation to the state of the art.