Annonce

★ The internship will take place within the FOX team of the CRIStAL laboratory at the University of Lille.

Advantage: This internship is supposed to be continued with a PhD thesis upon completion

Project Description : The project aims to design a neuromorphic system capable of detecting spatio-temporal patterns in real time within the activity of Spiking Neural Networks (SNNs). These networks process information through discrete events (spikes), relying on temporal coding, which enables sparse, energy-efficient computation.

A self-supervised learning approach allows the network to discover temporal regularities without manual annotation. The architecture will be implemented on FPGA hardware using the Address-Event Representation (AER) protocol, combining high speed, parallelism, and low power consumption.

The methodology includes:

• Software simulation using Brian2 and NEST
• Detection of polychronous patterns
• Optimized hardware transposition (fixed-point arithmetic, pipelining, compression).

The project will also explore deployment on neuromorphic hardware platforms such as Intel’s Loihi 2 [1] and SynSense Speck processors, known for their event-driven performance and energy efficiency.

This work aligns with the bio-inspired research axis of IRCICA, promoting frugal, adaptive brain-inspired architectures.

Research Topic and Scientific Context : The project “Pattern Extraction in Spiking Neural Networks and FPGA Implementation” lies within the field of neuromorphic computing, at the intersection of computational neuroscience, artificial intelligence, and reconfigurable hardware.

This rapidly growing field seeks to design brain-inspired systems capable of processing information in an asynchronous, distributed, and energy-efficient manner, as demonstrated by architectures such as Intel Loihi, IBM TrueNorth, and SpiNNaker (University of Manchester).

Spiking Neural Networks rely on Spike-Timing-Dependent Plasticity (STDP), a learning mechanism that enables the spontaneous emergence of spatio-temporal patterns reflecting internal network regularities [8]. These structures, analogous to primitive forms of memory, are central to understanding the link between local learning rules and neural pattern recognition.

While SNNs were initially applied to image classification, their potential for dynamic tasks—such as action recognition, optical flow, and medical imaging [3] has recently motivated research into self-supervised learning (SSL) for SNNs. However, the discrete nature of spikes makes it difficult to directly adapt SSL methods developed for conventional artificial neural networks.

Recent approaches, such as those proposed by Qiu et al. [4] and Singhal et al. [5], have begun to bridge this gap, though performance remains limited.

Project Objectives : The objective of this project is to develop a self-supervised framework for Spiking Neural Networks that exploits the temporal dynamics of spikes to extract recurrent patterns transferable to vision tasks (e.g., detection and segmentation).

These patterns will be implemented on FPGA hardware to demonstrate real-time detection with low energy consumption, extending the work of Barchid [8] on neural activity representation and hardware exploitation.

This research, at the intersection of self-supervised learning, event-based computation, and FPGA design, aims to lay the foundations for embedded neuromorphic systems capable of autonomous learning and adaptation, paving the way toward frugal, bio-inspired AI.

Scientific and Technological Objectives :

The main goal is to design, simulate, and implement a neuromorphic system capable of real-time neural pattern detection in Spiking Neural Networks. The project is structured around the following axes:

• Modeling and simulation of realistic SNNs using tools such as Brian2 or NEST, to generate spike streams representative of biological or synthetic neural activity.
• Development of a spatio-temporal pattern detection algorithm, inspired by statistical approaches such as SPADE (Spatio-Temporal Pattern Detection), based on identifying temporal coincidences between neurons.
• Integration of self-supervised learning: beyond classical STDP, the project will explore strategies allowing the network to use its own internal signals (correlations, rhythms, recurrences) as learning signals. The goal is to endow the system with autonomous adaptation capabilities, without labeled data.
• FPGA implementation using an event-driven architecture based on the AER protocol, requiring hardware-oriented optimizations such as:
  • Fixed-point arithmetic
  • Pipelining
  • Memory compression
  • Partial parallelization

Validation Methodology
The system will be validated through a rigorous two-phase methodology :

• Software phase:
  – Validation of pattern detection through simulation
  – Quantitative evaluation (precision, recall, F1-score)
• Hardware phase :
  – Performance measurement on FPGA (latency, power consumption, logic utilization)
  – Comparison between software and hardware implementations to assess efficiency and speed gains

Candidate Profile

• Final-year Master’s student (M2) or engineering school student
• Specialization in Machine Learning, Computer Vision, or related fields
• Knowledge of :
  • Computer vision
  • Machine learning and deep learning
• Programming skills in Python
• Strong autonomy, rigor, and critical thinking

★ The internship will take place within the FOX team of the CRIStAL laboratory at the University of Lille.

Internship Address:
CNRS – IRCICA-IRI-RMN
Campus Haute-Borne
Parc Scientifique de la Haute Borne
50 Avenue Halley, BP 70478
59658 Villeneuve d’Ascq Cedex, France

Application :

If you are interested, please send the following documents to Dr. Tanmoy Mondal (tanmoy.mondal@univ-lille.fr):

• Curriculum Vitae (CV)
• Motivation letter
• Academic transcripts (Bachelor’s / Master’s / Engineering degree) and class ranking
• Name and contact details of at least one academic or professional referee

Références

1. Mike Davies et al. Advancing neuromorphic computing with loihi : A survey of results and outlook. Proceedings of the IEEE, 109(5) :911–934, 2021.
   
2. Manish Darshan, Shih-Chii Zhang, and Shuang Liu. Speck : A general-purpose neuromorphic processor with multi-core on-chip learning support. In 2021 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 1–5. IEEE, 2021.
   
3. Alex Vicente-Sola, Davide L. Manna, Paul Kirkland, Gaetano Di Caterina, and Trevor J. Bihl. Spiking neural networks for event-based action recognition : A new task to understand their advantage. Neurocomputing, 611 :128657, 2025. ISSN 0925-2312. URL https://www.sciencedirect.com/science/article/pii/S0925231224014280.
   
4. Haonan Qiu, Zeyin Song, Yanqi Chen, Munan Ning, Wei Fang, Tao Sun, Zhengyu Ma, Li Yuan, and Yonghong Tian. Temporal contrastive learning for spiking neural networks. arXiv preprint arXiv :2305.13909, 2023
   
5. Raghav Singhal, Jan Finkbeiner, and Emre Neftci. Self-supervised pre-training of spiking neural networks by contrasting events and frames. In UniReps : 2nd Edition of the Workshop on Unifying Representations in Neural Models, 2024. URL https://openreview.net/forum?id=DNopfn4hZf.
   
6. Z. Zhou et al. Spikformer v2 : Join the high accuracy club on imagenet with an snn ticket. arXiv
   preprint arXiv :2401.02020, 2024b.
   
7. Jason K Eshraghian, Max Ward, Emre O Neftci, Xinxin Wang, Gregor Lenz, Girish Dwivedi, Mohammed Bennamoun, Doo Seok Jeong, and Wei D Lu. Training spiking neural networks using lessons from deep learning. Proceedings of the IEEE, 111(9) :1016–1054, 2023. doi :10.1109/JPROC.2023.3308088.
   
8. S. Barchid, Avancées en vision neuromorphique : représentation événementielle, réseaux de
   neurones impulsionnels supervisés et pré-entraînement auto-supervisé, Thèse de doctorat,
   Université de Lille, 2023.