Allen Institute for Neural Dynamics’ Post

If you are attending NeurIPS 2023 in New Orleans this week, please check out our work: 1. "Every Parameter Matters: Ensuring the Convergence of Federated Learning with Dynamic Heterogeneous Models Reduction". https://lnkd.in/gisWzG6y 2. "A Unified Algorithm Framework for Unsupervised Discovery of Skills based on Determinantal Point Process". https://lnkd.in/gtZfyD_U

Earlier this month, I had the opportunity to present my work at the 21st International Symposium on Applications of Laser and Imaging Techniques and at the workshop on the Future Perspectives of Defocus Particle Tracking in Lisbon. We provide a #DeepLearning workflow for predicting diffusion coefficient from particle images that work even if there is a flow or the particles are defocused. More broadly, our work provides the #fluids community #ResNet styled image and video regression #CNN architectures optimized for particle images and can serve as better #checkpoints for other applications such as #PIV and #uPIV. Our #OpenSource code will also contain our #NeuralArchitectureSearch and #HyperparameterOptimization scripts to make it easier to optimize the architectures for other applications. In the meantime, you can learn more about our work from our previous work: https://lnkd.in/gJ4fqMap #LXLASER2024 #Lisbon2024 #particleimagevelocimetry #Velocimetry #PIV #CNNs #ConvolutionalNeuralNetworks #ImageRegression #VideoRegression

Delving into YOLOv10n: Object Detection Powerhouse [Part 1] Excited to kick off a 3-part series exploring the capabilities of YOLOv10, the latest object detection marvel from Ultralytics!  We're starting small but mighty, training YOLOv10 on a blood cell detection dataset (https://lnkd.in/gmpG9a7J). The results? Impressive!  We test the trained YOLOv10 on a random blood cell video found online. It's pinpointing those red and white blood cells with impressive accuracy!  Want to replicate the magic? Grab the YOLOv10 code from the official Ultralytics (https://lnkd.in/gRyPsWqg) Stay tuned for Part 2, where we'll crank up the complexity and see how YOLOv10 handles a real-world object detection challenge!  #YOLOv10 #Ultralytics #ObjectDetection #MachineLearning #BloodCellDetection

We think Visium #CytAssist is essential to generating best-in-class spatial data with the #VisiumHD assay and to helping you achieve your best spatial experiment yet. Explore 4 reasons why in this post. Learn more at: https://bit.ly/49lzJTp

Did you know IEEE Brain has a free archive of #webinars just waiting for you to view? Each episode features leading experts sharing their knowledge and expertise into topics like #brain-machine interfaces, #neuromorphic systems, #DeepBrainStimulation, and more. Tune in now at: https://bit.ly/3KKXBV6 #BCI #BMI #DBD #neurotech

My paper “Characterizing the structure of mouse behavior using Motion Sequencing” is the featured protocol of the week on Nautre Protocols homepage. This protocol offers a thorough instructions and the code for the MoSeq pipeline, an ethologically inspired method that uses three-dimensional machine vision and unsupervised machine learning to parse animal naturalistic behavior into modulated, repeated motifs. This protocol is one of many things my lab and my team have done to make the MoSeq platform more user friendly and accessible. Check out the paper here:

Our paper was accepted at NeurIPS 2023 Workshop on Diffusion Models! 🎉 This time, we extended our work for historical maps! 🏛️🗺️ We train on UK's 1888 map to generate satellite images that follow the map layout. Paper and code: https://lnkd.in/eV9B3cN3 Note that we train on a "semi-paired" dataset, as the real satellite images (~2017) do not match entirely with the historical UK map (1888) due to the time difference (more than 100 years!). Yet, the generated satellite images are consistent with the map.

🔍 Exploring Atrous Convolutions in CNNs: Unlocking Image Details! 🌟 Delve into the fascinating world of Atrous Convolutions with my latest blog! From understanding how CNNs analyze intricate images to the magic of Dilated Convolutions for multi-scale feature learning, this blog takes you on a journey through the structure, applications, and famous architectures using Atrous Convolution. Check out the blog for a deeper dive into: 🔸How Atrous Convolutions add clarity to complex images 🖼️ 🔸Unveiling famous architectures like DeepLab, LinkNet, and more 🏗️ 🔸Key takeaways and FAQs to grasp the essence of Atrous Convolution 🧠 Read the full blog here : https://lnkd.in/d-vTEx2c #AtrousConvolution #CNNs #ImageProcessing #DeepLearning

Key takeaways from PUZZLE X™ by The Innovator Quantum vs AI = which one has a greener environmental impact on the planer? Special thanks to Román Orús, Chief Scientific Officer at Multiverse Computing for his inspiring and thought provoking showcase: "AI is burning the planet faster than airplanes" & "“We need a new method of computing in order to make it greener and more sustainable” Astonishingly, the cost of training a single AI model like OpenAI's GPT-4 surpassed US$100 million in electricity bills. An estimation suggests that LLM consumed as much electricity as 175,000 humans in one month.

As a postdoc back in 2004, I enjoyed an early morning coffee chat with the inimitable Professor Matthew Wilson (MIT), speculating about hippocampal place cells and theta phase coding. One thing, eventually, led to another: our preprint 'Integration of rate and phase codes by hippocampal cell-assemblies supports flexible encoding of spatiotemporal context' https://lnkd.in/e9qvmmDX