NeuroAI & Machine Learning | Laurent Perrinet

Al for visual processing and energy-efficient algorithms

Tue, 03 Feb 2026 00:00:00 +0000

A New Look for Convolutional Deep Networks

Fri, 12 Dec 2025 00:00:00 +0000

🔬 Excited to present our latest research at the #MAIN2025 conference today!

🔗 https://www.main2025.org/

👁️ What if CNNs could see like humans? Our new work shows how foveated vision—concentrating processing at gaze center—makes networks more robust to perturbations & great at localization. Inspired by human vision’s architecture (high-resolution foveal center, low-resolution periphery), we embedded this retinotopic transformation into CNN architectures, allowing to actively scan the image. This gives literally a new look to #ConvNets !

📄 Paper:

Jean-Nicolas Jérémie, Emmanuel Daucé, Laurent U Perrinet (2026). Foveated Retinotopy Improves Classification and Localization in CNNs. Vision.

PDF Cite DOI URL arXiv

#DeepLearning #ComputerVision #AI #Research #NeuralNetworks #NeuroAI #OpenScience I love #Montreal

Media storm - share if you like :

A New Look for Convolutional Deep Networks

Tue, 02 Dec 2025 00:00:00 +0000

Mislocalization by Design: The Flash-Lag Effect as Prediction

Thu, 16 Oct 2025 00:00:00 +0000

Qu'est-ce que les Neurosciences peuvent apporter à l'Intelligence Artificielle ?

Fri, 14 Feb 2025 00:00:00 +0000

When Cortical Neurons Talk Sideways: Beyond Feedforward Visual Processing

Tue, 11 Feb 2025 00:00:00 +0000

In this seminar we will challenge the traditional understanding of neuronal connectivity in primary visual cortex. While current theory suggests that neurons connect preferentially to others with similar orientation preferences, I will present evidence for a more complex connectivity pattern based on a distance-dependent rule: short-range connections show a like-to-like bias, while long-range connections connect more widely. This revised model better explains how the visual cortex processes complex stimuli and accounts for observed variations in neuronal interactions at different scales.

When: Wednesday 11th of February, 2025 from 14:30 to 16h30.
Where: room D2.2 of Collège de France
Accompanying paper
Revisiting Horizontal Connectivity Rules in V1: From like-to-like towards like-to-All
Frédéric Chavane, Laurent U Perrinet, James Rankin

PDF Cite DOI URL

How and why foveated retinotopy provides efficient vision

Wed, 08 Jan 2025 00:00:00 +0000

When: Wednesday 9th of January, 2025 at 12 noon.
Where: CRN seminar room, Montreal General Hospital, Livingston Hall, L7-140, with hybrid option.
Read the corresponding paper
Foveated Retinotopy Improves Classification and Localization in CNNs
Jean-Nicolas Jérémie, Emmanuel Daucé, Laurent U Perrinet

PDF Cite DOI URL arXiv

Integrating the What and Where Visual Pathways to Improve CNN Categorisation

Wed, 01 Jan 2025 00:00:00 +0000

🧠 TODAY at #CCN2025 ! Poster A145, 1:30-4:30pm at de Brug & E‑Hall. We’ve developed a bio-inspired “What-Where” CNN that mimics primate visual pathways - achieving better classification with less computation. Come chat! 🎯

What: Poster A145
When: Tuesday, August 12, 1:30 – 4:30 pm,
Where: CCN 2025 conference venue, de Brug & E‑Hall

Presented by main author Jean-Nicolas JÉRÉMIE and in cosupervision with Emmanuel Daucé

https://laurentperrinet.github.io/publication/jeremie-25-ccn/

Our research introduces a novel “What-Where” approach to CNN categorization, inspired by the dual pathways of the primate visual system:

The ventral “What” pathway for object recognition
The dorsal “Where” pathway for spatial localization

Key innovations:

✅ Bio-inspired selective attention mechanism

✅ Improved classification performance with reduced computational cost

✅ Smart visual sensor that samples only relevant image regions

✅ Likelihood mapping for targeted processing

The results?

Better accuracy while using fewer resources - proving that nature’s designs can still teach us valuable lessons about efficient AI.

Come find us this afternoon for great discussions!

#CCN2025 #ComputationalNeuroscience #AI #MachineLearning #BioinspiredAI #ComputerVision #Research

Links:

NeuroAI: interactions multiples entre Neurosciences et Intelligence artificielle

Mon, 18 Nov 2024 00:00:00 +0000

présentation de 15:50:00 à 16:20:00

A Robust Event-Driven Approach to Always-on Object Recognition

Tue, 01 Oct 2024 00:00:00 +0000

Main contributions:

Builds an adaptive, back to back event-based pattern recognition architecture, inspired by neuroscience and capable of always-on decision, that is, that the decision can be taken it can be needed,

The HOTS architecture.
Extends the HOTS algorithm to increase its performance by adding a homeostatic gain control on the activity of neurons to improve the learning of spatio-temporal patterns, we prove an analogy with off-the-shelf LIF spiking neurons,
Quantitative results assessing the role of homeostasis in the learning of spatio-temporal patterns and the performance of the network on different datasets: Poker-DVS, N-MNIST and DVS Gesture for different precisions of the temporal and spatial information.

Performance of the algorithm on the DVSgesture dataset. For this gesture recognition task, the online HOTS accuracy remains close to the chance level for about 100 events. More evidence needs to be accumulated, and then the accuracy increases monotonically, outperforming the previous method after about 10.000 events (at an average of 9.3% of the number of events in the sample).

Robots aériens agiles bio-mimetiques pour le vol en conditions réelles: Enjeux pour l'IA embarqué

Mon, 09 Sep 2024 16:20:00 +0000

Time-to-Contact Map by Joint Estimation of Up-to-Scale Inverse Depth and Global Motion using a Single Event Camera

Fri, 06 Oct 2023 00:00:00 +0000

the code is openly available on GitHub with the accompanying data VL.zip.

Learning heterogeneous delays in a layer of spiking neurons for fast motion detection

Mon, 11 Sep 2023 00:00:00 +0000

read the paper online (paywall) or read the reprint as PDF
full code with extensive Supplementary Material
join the Zotero group to add and discuss more items
this paper is a follow-up of
Antoine Grimaldi, Camille Besnainou, Hugo Ladret, Laurent U Perrinet (2022). Learning heterogeneous delays of spiking neurons for motion detection. Proceedings of ICIP 2022.
PDF Cite DOI Venue URL

Stakes of Neuromorphic Foveation: a promising future for embedded event cameras

Sun, 01 Jan 2023 00:00:00 +0000

Learning heterogeneous delays of spiking neurons for motion detection

Sun, 16 Oct 2022 00:00:00 +0000

see a follow-up as journal paper:
Antoine Grimaldi, Laurent U Perrinet (2023). Learning heterogeneous delays in a layer of spiking neurons for fast motion detection. Biological Cybernetics.
PDF Cite DOI URL Code Supplementary Material
presented at ICIP 2022 16-19 October 2022 in Bordeaux, France
paper 3241 (note that the title of the paper was slightly changed)
time of presentation:
Tue, 18 Oct, 20:30 - 20:45 China Standard Time (UTC +8)
Tue, 18 Oct, 14:30 - 14:45 Central European Time (UTC +2)
Tue, 18 Oct, 12:30 - 12:45 UTC
Tue, 18 Oct, 08:30 - 08:45 Eastern Time (UTC -4)

Session “Neuromorphic and perception-based image acquisition and analysis”

TQ-L.A Special session on Tueasday, October 18 from 14:00 to 16:00
Organized by Dr. Marc Antonini, Dr. Panagiotis Tsakalides, and Dr. Effrosyni Doutsi:

During the last decade much attention has been paid to understanding the human brain properties and functions in order to mimic the computational mechanisms of this highly intelligent processing “machine” that seems to be able to address several technological challenges that the scientific community is currently facing. Digital sobriety is quite important among these challenges as it concerns the reduction of the energy footprint caused by the use and transmission of the digital information. According to recent studies, almost 80% of global data flows is due to online videos stored in big data centers ready to be accessed on demand at any time by several users all over the world. As a result, scientists are urged to find energy-saving solutions to capture, process, understand, compress and stream this great volume of visual information in an environmental responsible and greener manner. Brain-inspired or neuro-inspired or spike-based or event-based computing are all terms used to describe the emerging technological trend motivated by the brain capability to dynamically capture and to spatio-temporally process and transform the great volume of the 3D visual information into a very compact spike train that is fed forward to the visual cortex of the brain passing through a very dense neural network. This is an energy efficient process, a fact that triggered the attention of the signal processing community trying to design more sober video services. Indeed, every step of the brain processing pipeline provides inspiration towards novel disruptive implementations of image and video processing components: (i) visual sensors responsible for capturing and projecting the visual information into a neuromorphic chip, (ii) image understanding utilizing spiking neural networks to better approximate the dense interconnected network of neurons along the visual pathway, (iii) image processing and compression motivated by the exceptional compactness of the spike trains, capable of providing an ultra-high-definition perception of the visual world. In addition, the last decade has witnessed the progress of neuromorphic algorithms and hardware, which has already reached performance and manufacturing levels that is beyond the state- of-the-art. The objective of this special session is to highlight the importance of neuromorphic computing in image and video processing. We are interested in bringing together scientists working on different spike-based computational models, from sensing to understanding, who will share their knowledge and discuss about the advantages and the limitations of this type of systems. The aim is to progress towards an end-to-end and robust technology where the hardware and software will both follow the same neuro-inspired principles, addressing important challenges of the current conventional systems. Last but not least, this special session would be a great opportunity to build a strong international consortium between different teams to attract European and international funding to further study and promote neuromorphic computing for different signal processing open challenges.

Pooling in a predictive model of V1 explains functional and structural diversity across species

Mon, 18 Jul 2022 00:00:00 +0000

this paper follows this COSYNE presentation :
Angelo Franciosini, Victor Boutin, Laurent U Perrinet (2020). Modelling Complex-cells and topological structure in the visual cortex of mammals using Sparse Predictive Coding. Computational and Systems Neuroscience (Cosyne) 2020.
Cite URL
see a related work describing SDPC in:
Victor Boutin, Angelo Franciosini, Frédéric Chavane, Franck Ruffier, Laurent U Perrinet (2021). Sparse Deep Predictive Coding captures contour integration capabilities of the early visual system. PLoS Computational Biology.
PDF Cite DOI Code URL arXiv

more about the role of top-down connections:
Victor Boutin, Angelo Franciosini, Franck Ruffier, Laurent U Perrinet (2020). Effect of top-down connections in Hierarchical Sparse Coding. Neural Computation.
PDF Cite DOI URL arXiv

Learning heterogeneous delays of spiking neurons for motion detection

Tue, 12 Jul 2022 00:00:00 +0000

for a follow-up, check out
Antoine Grimaldi, Camille Besnainou, Hugo Ladret, Laurent U Perrinet (2022). Learning heterogeneous delays of spiking neurons for motion detection. Proceedings of ICIP 2022.
PDF Cite DOI Venue URL

Decoding spiking motifs using neurons with heterogeneous delays

Wed, 29 Jun 2022 00:00:00 +0000

for a follow-up, check out
Antoine Grimaldi, Camille Besnainou, Hugo Ladret, Laurent U Perrinet (2022). Learning heterogeneous delays of spiking neurons for motion detection. Proceedings of ICIP 2022.
PDF Cite DOI Venue URL
see a follow-up as journal paper:
Antoine Grimaldi, Laurent U Perrinet (2023). Learning heterogeneous delays in a layer of spiking neurons for fast motion detection. Biological Cybernetics.
PDF Cite DOI URL Code Supplementary Material

Learning heterogeneous delays of Spiking Neurons for motion detection

Sun, 19 Jun 2022 09:00:00 +0000

for a follow-up, check out
Learning heterogeneous delays of spiking neurons for motion detection
Antoine Grimaldi, Camille Besnainou, Hugo Ladret, Laurent U Perrinet

PDF Cite DOI Venue URL

Contributions of neuroscience to the detection and localization of objects in visual inputs

Tue, 14 Jun 2022 15:00:00 +0000

for visual search see:
A dual foveal-peripheral visual processing model implements efficient saccade selection
Emmanuel Daucé, Pierre Albigès, Laurent U Perrinet

Cite DOI Pdf Code URL bioRxiv
for retinotopy, see:
Retinotopic mapping improves the reliability of image classification
PDF Cite URL
for event-based computations, see:
Learning heterogeneous delays of Spiking Neurons for motion detection
PDF Cite URL
for event-based motion detection, see:
Learning heterogeneous delays of spiking neurons for motion detection
Antoine Grimaldi, Camille Besnainou, Hugo Ladret, Laurent U Perrinet

PDF Cite DOI Venue URL

Polychrony detection using heterogeneous delays

Thu, 19 May 2022 09:00:00 +0000

Follow this future presentations
Learning heterogeneous delays of Spiking Neurons for motion detection
PDF Cite URL
followed-up as a poster:
Decoding spiking motifs using neurons with heterogeneous delays
Antoine Grimaldi, Camille Besnainou, Hugo Ladret, Laurent U Perrinet

PDF Cite Slides Venue
for event-based motion detection, see:
Learning heterogeneous delays of spiking neurons for motion detection
Antoine Grimaldi, Camille Besnainou, Hugo Ladret, Laurent U Perrinet

PDF Cite DOI Venue URL

Ultra-fast categorization of images containing animals in vivo and in computo

Fri, 15 Oct 2021 00:00:00 +0000

see a follow-up in:
Jean-Nicolas Jérémie, Laurent U Perrinet (2023). Ultra-Fast Image Categorization in biology and in neural models. Vision.
Cite DOI arXiv
see an extension to visual search in:
Jean-Nicolas Jérémie, Emmanuel Daucé, Laurent U Perrinet (2022). Ultra-rapid visual search in natural images using active deep learning. Proceedings of AREADNE.
PDF Cite Venue

From event-based computations to a bio-plausible Spiking Neural Network

Thu, 14 Oct 2021 00:00:00 +0000

this proceedings paper follows up the poster presented at CBMI :
Antoine Grimaldi, Victor Boutin, Sio-Hoi Ieng, Laurent U Perrinet, Ryad Benosman (2021). A homeostatic gain control mechanism to improve event-driven object recognition. Content-Based Multimedia Indexing (CBMI) 2021.
PDF Cite DOI URL HAL Video
read the follow-up paper :
Antoine Grimaldi, Victor Boutin, Sio-Hoi Ieng, Ryad Benosman, Laurent U Perrinet (2024). A Robust Event-Driven Approach to Always-on Object Recognition. Neural Networks.
PDF Cite DOI Press Hal Code URL

A homeostatic gain control mechanism to improve event-driven object recognition

Thu, 24 Jun 2021 00:00:00 +0000

was presented at the Bio-inspired circuits, systems and algorithms for multimedia special session of the Content-Based Multimedia Indexing (CBMI) 2021 conference that you can watch on Youtube.
this proceedings paper follows up he poster presented in :
Antoine Grimaldi, Victor Boutin, Sio-Hoi Ieng, Laurent U Perrinet, Ryad Benosman (2021). A robust bio-inspired approach to event-driven object recognition. Computational and Systems Neuroscience (Cosyne) 2021.
PDF Cite Pdf URL
this proceedings paper was followed by the poster presented at CRS :
Antoine Grimaldi, Victor Boutin, Sio-Hoi Ieng, Ryad Benosman, Laurent U Perrinet (2021). From event-based computations to a bio-plausible Spiking Neural Network. Champalimaud Research Symposium (CRS21).
PDF Cite Video Venue
read the follow-up paper :
Antoine Grimaldi, Victor Boutin, Sio-Hoi Ieng, Ryad Benosman, Laurent U Perrinet (2024). A Robust Event-Driven Approach to Always-on Object Recognition. Neural Networks.
PDF Cite DOI Press Hal Code URL
Antoine Grimaldi and Laurent Perrinet received funding from the European Union ERA-NET CHIST-ERA 2018 research and innovation program under grant agreement No ANR-19-CHR3-0008-03.

A robust bio-inspired approach to event-driven object recognition

Fri, 26 Feb 2021 00:00:00 +0000

see the poster online on the Hopin platform
see a follow-up in:
Antoine Grimaldi, Victor Boutin, Sio-Hoi Ieng, Laurent U Perrinet, Ryad Benosman (2021). A homeostatic gain control mechanism to improve event-driven object recognition. Content-Based Multimedia Indexing (CBMI) 2021.
PDF Cite DOI URL HAL Video
read also the follow-up paper :
Antoine Grimaldi, Victor Boutin, Sio-Hoi Ieng, Ryad Benosman, Laurent U Perrinet (2024). A Robust Event-Driven Approach to Always-on Object Recognition. Neural Networks.
PDF Cite DOI Press Hal Code URL
Antoine Grimaldi and Laurent Perrinet received funding from the European Union ERA-NET CHIST-ERA 2018 research and innovation program under grant agreement No ANR-19-CHR3-0008-03.

Sparse Deep Predictive Coding captures contour integration capabilities of the early visual system

Tue, 26 Jan 2021 00:00:00 +0000

Fig 1. Architecture of a 2-layered SDPC model.

One often compares biological vision to a camera-like system where an image would be processed according to a sequence of successive transformations. In particular, this “feedforward” view is prevalent in models of visual processing such as deep learning. However, neuroscientists have long stressed that more complex information flow is necessary to reach natural vision efficiency. In particular, recurrent and feedback connections in the visual cortex allow to integrate contextual information in our representation of visual stimuli. These modulations have been observed both at the low-level of neural activity and at the higher level of perception.

Fig 2. Results of training SDPC on the natural images (left column) and on the face database (right column) with a feedback strength kFB = 1.

Fig 14. Illustration of the hierarchical generative model learned by the SDPC model on the face database.

In this study, we present an architecture that describes biological vision at both levels of analysis. It suggests that the brain uses feedforward and feedback connections to compare the sensory stimulus with its own internal representation. In contrast to classical deep learning approaches, we show that our model learns interpretable features.

Fig 5. Example of a 9 × 9 interaction map of a V1 area centered on neurons strongly responding to a central preferred orientation of 30°.

Fig 7. Example of a 9 × 9 interaction map of a V1 area centered on neurons strongly responding to a central preferred orientation of 45°, and colored with the relative response w.r.t. no feedback.

Moreover, we demonstrate that feedback signals modulate neural activity to promote good continuity of contours. Finally, the same model can disambiguate images corrupted by noise. To the best of our knowledge, this is the first time that the same model describes the effect of recurrent and feedback modulations at both neural and representational levels.

Fig 10. Effect of the feedback strength on noisy images from natural images database.