Martine Bertrand

There's only 5 days left to submit your paper! Remember, we only a short 1-page abstract

2

Last week, Alexandre Dubray presented "Anytime Weighted Model Counting with Approximation Guarantees for Probabilistic Inference", joined work with Pierre Schaus, at the Constraint Programming conference. In this work, we show how to modify a model counting solver to calculate an approximate model count, with approximation guarantees at any moment the solver is stopped. We believe such a solver can be useful for inference tasks on many probabilistic models: Bayesian networks, neurosymbolic models, and more. https://lnkd.in/eg8FQcBc

21

John J. Hopfield of Princeton University and Geoffrey E. Hinton of the University of Toronto have won the 2024 #NobelPrize in Physics for machine learning research. They share the prize for “foundational discoveries and inventions that enable machine learning with artificial neural networks,” a technology that was inspired by the structure of the brain. This story will be updated! https://lnkd.in/ew_wWHwy

52

We have figured out a lot using mathematics, physics and chemistry to understand nature. We know how to predict electron flow in semiconductors, fluids in tubes, clouds in the sky, airflow over wings, space storms etc. Tools like computational fluid dynamics have remarkable applications in unexpected places. We also have ideas around PINN and we can use those insights to help us understand what happens in a foundational model. And I am sure there are lots more. None of these software artifacts should ever be said to be un-knowable or a "blackbox". That's intellectual laziness. These are all man made structures. We've figured out many parts of nature using math, physics and chemistry - we've mapped the Human Genome, so what's the excuse in not understanding foundational models? I look forward to people recognizing that there are existing human inventions (math, physics etc) whose insights can be used to remove/reduce the probabilistic nature of foundational models and thereby benefiting humanity as a whole. A press conference was held by University of Toronto with Geoffrey Hinton after the announcement that he was one of the two recipients of this year's Nobel Prize for Physics. Geoffrey is probably the only Turing Award winner to also have won a Nobel Prize. And the video - https://lnkd.in/gVHewNCd. Geoffrey is spot on where he says, at about 3:38: "I'm particularly proud of the fact that one of my students fired Sam Altman." Brilliant! #nobelprize #ai #opensource #turingaward

6

3 Comments

Inverse rendering is an ill-posed problem as many parameters have to be estimated like lighting, materials, geometry, camera parameters (poses, calibration, lens distortions). Diffusion models to the rescue! Presenting "MaterialFusion: Enhancing Inverse Rendering with Material Diffusion Priors" from Carnegie Mellon University and Tel Aviv University. They present StableMaterial, a 2D diffusion model prior that refines multi-lit data to estimate the most likely albedo and material from given input appearances. This model is trained on albedo, material, and relit image data derived from a curated dataset of approximately ~12K artist-designed synthetic Blender objects called BlenderVault. They incorporate this diffusion prior with an inverse rendering framework where they use score distillation sampling (SDS) to guide the optimization of the albedo and materials, improving relighting performance in comparison with previous work. They validate MaterialFusion's relighting performance on 4 datasets of synthetic and real objects under diverse illumination conditions, showing their diffusion-aided approach significantly improves the appearance of reconstructed objects under novel lighting conditions. Checkout all the relevant links in the comments #inverserendering #materialestimation #diffusionmodels #generativeai #vfx #reconstruction #sceneunderstanding #materialunderstanding #machinelearning #deeplearning #futurism #computervision

68

1 Comment

AI for Science: a paradigm shift for scientific discovery and translation: "From the landmark US CHIPS and Science Act legislation to major initiatives such as Canada’s Acceleration Consortium, leaders around the world are placing big investments in scientific research and innovation to solve today’s grand challenges and those yet to come. To meet the growing urgency of our time, the fundamental way scientific research is conducted in the future must be transformed by advancing and connecting recent breakthrough technologies across the physical and virtual worlds — from remotely controlled, high-throughput automated experimental laboratories at scale all the way to new and powerful foundational AI and physics-based models driven by high-performance computing and rich data resources not previously available." https://lnkd.in/gCdmcfdv

13

scGPT: A New Model Based on Generative AI in the Field of Biology Introducing scGPT, an innovative model developed by researchers at the University of Toronto and the Vector Institute, designed to revolutionize the way we understand single-cell multi-omics. By harnessing the power of generative AI, scGPT is trained on over 10 million single-cell datasets, enabling it to uncover profound biological insights at both cellular and genetic levels. scGPT is a game-changer in tasks like cell type annotation, genetic perturbation prediction, and gene network inference. This model is not just a tool; it's a leap forward in cellular biology research, paving the way for groundbreaking discoveries. scGPT is set to transform the future of single-cell analysis. I added the link to the full paper, which I read for your convenience: https://lnkd.in/esYCQV_B #AI #MachineLearning #Bioinformatics #GenerativeAI #SingleCellOmics #scGPT #Biotechnology #ResearchInnovation #Transformers #CellBiology #DataScience #Omics #GenomeResearch #AIInHealthcare #Science #Technology #Innovation #DeepLearning #BiomedicalResearch #AIRevolution #FutureOfScience

3

A question to the community: Do we need to include diffusion inside the model's architecture as done by NVIDIA's EDM paper for UNET, or just tinkering with the input data should be okay ? What I mean is adding noise in time steps and teaching the model to map from noisy data to the target in steps is okay ? I want to add the diffusion as in NVIDIA's EDM work to all models from SRCNN to Swin Transformer to IPG to SRGAN ... For example in the EDM networks.py code at https://lnkd.in/gUJ2azPk for DhariwalUNET:   def forward(self, x, noise_labels, class_labels, augment_labels=None):     # Mapping.     emb = self.map_noise(noise_labels)     if self.map_augment is not None and augment_labels is not None:       emb = emb + self.map_augment(augment_labels)     emb = silu(self.map_layer0(emb))     emb = self.map_layer1(emb)     if self.map_label is not None:       tmp = class_labels       if self.training and self.label_dropout:         tmp = tmp * (torch.rand([x.shape[0], 1], device=x.device) >= self.label_dropout).to(tmp.dtype)       emb = emb + self.map_label(tmp)     emb = silu(emb)     # Encoder.     skips = []     for block in self.enc.values():       x = block(x, emb) if isinstance(block, UNetBlock) else block(x)       skips.append(x)     # Decoder.     for block in self.dec.values():       if x.shape[1] != block.in_channels:         x = torch.cat([x, skips.pop()], dim=1)       x = block(x, emb)     x = self.out_conv(silu(self.out_norm(x)))     return x the map_noise is used even in the forward function. I wanted to use diffusion for models like Swin Transformer and IPG and changing within the model architecture seems challenging. Background: Recently I have been dabbling a lot with the diffusion models, and now am interested to add diffusion to all image to image models from SRCNN to Swin Transformers. The inspiration has spiked from Peter Manshausen's work on diffusion based data assimilation. Also, there is a very nice paper from NVIDIA titled "Elucidating the Design Space of Diffusion-Based Generative Models" by Tero Karras and others. It provides clear design choices for diffusion models.

52

Amazing research on controllable character video synthesis via generative diffusion models. Here, they introduce MIMO, a generalizable model that is capable of synthesizing character videos with controllable attributes (e.g., character, motion, scene, etc.), allowing for great flexibility. They achieve this through latent codes that encapsulates motion, scene, pose, character identity, etc. that are inserted into a diffusion decoder as conditions as part of the video generation process. #diffusion #generativeai #genai #generative #ai #artificialintelligence #ml #machinelearning #deeplearning #dl #video #character #synthesis #cool #research #control #character #motion #scene

41

3 Comments

Very pleased to report that my article with Hunter Gentry, "Transitional gradation and the distinction between episodic and semantic memory", has just appeared in _Philosophical Transactions of the Royal Society B". It is also open access thanks to the new resources I have available at University of Florida, specifically an OA agreement with the UF library and the Royal Society. So everybody can see it without any shenanigans! In this article we explore various arguments against Tulving's traditional episodic/semantic memory distinction, provide them on a firmer metaphysical foundation by drawing upon the theory of natural kinds and previous things I've said about a phenomenon called "transitional gradation" in the special sciences, recommend replacing the distinction with a more dynamic life cycle of mnemonic forms of representation, and sketch some new ideas for experiments inspired by the perspective that can help arbitrate the debate over episodic memory in animals (largely Hunter's work here). Hunter was a joy to work with on this and if you see an application from him you should give it serious consideration. He's the real deal and will be publishing stuff like this for decades to come. https://lnkd.in/gPMJSV7Q

14

NVIDIA Research is pushing the boundaries of AI and simulation at SIGGRAPH 2024! 🚀 Key highlights: • 20+ papers introducing innovations in synthetic data generation, inverse rendering, and AI-powered rendering • Advancements in diffusion models for visual generative AI, including faster consistent character generation and interactive texture painting • Breakthroughs in physics-based simulation, including text-to-motion generation and neural physics for various 3D representations • New rendering techniques for improved realism and diffraction simulation, with up to 1000x speedups • Novel AI tools for 3D representations and design, including a GPU-optimized framework for large-scale 3D deep learning These developments will empower creators, scientists, and engineers to generate complex virtual worlds, improve simulations, and push the boundaries of computer graphics. Don't miss Jensen Huang's fireside chats with Mark Zuckerberg and WIRED's Lauren Goode at the event! #SIGGRAPH2024 #NVIDIA #AIResearch #ComputerGraphics #Simulation

8

Bringing together the best of AI and antibody modeling to make those complex predictions a little less... complex. Just published from the Molecular Modeling team at the National Research Council Canada / Conseil national de recherches Canada, “AI-Augmented Physics-Based Docking for Antibody-Antigen Complex Prediction”. We present an AI-enhanced approach to improve antibody docking predictions, a key step toward designing better antibody therapeutics. This study shows how AI-guided structural diversity and smart model selection significantly boost docking accuracy, outperforming conventional methods. Thank you to my co-authors, Francis Gaudreault and Traian Sulea. Read more in the full paper! https://lnkd.in/gjDupeqC #AI #AntibodyResearch #AlphaFold #MachineLearning #Innovation #Antibods #docking

51

1 Comment

🌟 Exciting News from the Canadian Space Agency | Agence spatiale canadienne🚀 I had the distinct honor of presenting at the Canadian Space Agency | Agence spatiale canadienne on a topic that is not only fascinating but also critical to the future of space travel: "Exploring the impact of Nutri-omics on enhancing astronaut health and performance during space travel considering Health Beyond Initiative project." 🔬 In my talk, I delved into how Nutri-Omics — a pioneering field that merges personalized nutrition with advanced "omics" technologies — can revolutionize our understanding of the molecular interactions between food and human health. By leveraging "nutrigenomics" and cutting-edge "Machine learning techniques", we are paving the way for customized dietary strategies that cater specifically to the unique physiological needs of astronauts during their missions. 🧬 This comprehensive approach allows us to predict and manage health issues that might arise in the challenging environment of space, ensuring better health outcomes and performance optimization for astronauts on long-duration missions. 🤖 Machine learning, a crucial component of our research, helps us handle complex data and extract meaningful insights which are essential in predicting personalized nutrition plans. I am grateful to have had the opportunity to share my insights and findings with a keen audience at the Canadian Space Agency | Agence spatiale canadienne, contributing to the "Health Beyond Initiative Project." A huge thanks to dedicated team at Canadian Space Agency | Agence spatiale canadienne especially Annie Martin (Health Beyond Portfolio Manager at Canadian Space Agency) and Gurneet Jassal (Space Health operation assistant. https://lnkd.in/exxwpg_C 🌌 As we continue to explore the vast unknowns of space, it's initiatives like these that ensure our astronauts are not only safer but healthier and better equipped to handle the rigors of space travel. #SpaceNutrition #NutriOmics #AstronautHealth #MachineLearning #SpaceTravel #CanadianSpaceAgency #InnovationInSpace

58

4 Comments

This collaborative project did it: leveraging single-photon sensing technology to develop new brain neuromonitoring systems ➡️ real time (x100 speed) ⏲️, at 1/10th the cost of a PET scan💲. NRC's Quantum Sensors Challenge Program has an explicit challenge statement "to develop a disruptive generation of quantum sensors that are orders of magnitude better than sensors that exist today". What's also exciting is that this isn't the only type of project in our 🇨🇦 program taking these strides: quantum tech is moving out of the dark quantum labs and into sensing application spaces. #NRCChallengeQuantumSensors #QuantumTechnology #Quantum #QuantumSensing #NationalQuantumStrategy

35

🚨ECCV 2024 Paper Alert 🚨 ➡️Paper Title: Volumetric Rendering with Baked Quadrature Fields 🌟Few pointers from the paper 🎯In this paper authors have proposed a novel Neural Radiance Field (NeRF) representation for non-opaque scenes that enables fast inference by utilizing textured polygons. 🎯Despite the high-quality novel view rendering that NeRF provides, a critical limitation is that it relies on volume rendering that can be computationally expensive and does not utilize the advancements in modern graphics hardware. 🎯Many existing methods fall short when it comes to modeling volumetric effects as they rely purely on surface rendering. They thus propose to model the scene with polygons, which can then be used to obtain the quadrature points required to model volumetric effects, and also their opacity and color from the texture. 🎯To obtain such polygonal mesh, they trained a specialized field whose zero-crossings would correspond to the quadrature points when volume rendering, and perform marching cubes on this field. They then perform ray-tracing and utilize the ray-tracing shader to obtain the final color image. 🎯Their method allows an easy integration with existing graphics frameworks allowing rendering speed of over 100 frames-per-second for a 1920x1080 image, while still being able to represent non-opaque objects. 🏢Organization: The University of British Columbia, Google DeepMind, University of Toronto, Simon Fraser University 🧙Paper Authors:  Gopal Sharma, Daniel Rebain, Kwang Moo Yi, Andrea Tagliasacchi 1️⃣Read the Full Paper here: https://lnkd.in/gBwXNAeX 2️⃣Project Page: https://lnkd.in/gRaZwjrp 3️⃣Code: https://lnkd.in/gGg8yREi 🎥 Be sure to watch the attached Demo Video -Sound on 🔊🔊 Find this Valuable 💎 ? ♻️REPOST and teach your network something new Follow me 👣, Naveen Manwani, for the latest updates on Tech and AI-related news, insightful research papers, and exciting announcements. #ECCV2024 #NERF

18

The process of computing vegetation indexes in the plimanshiny app. Begins by navigating to the "Index" section after importing your mosaic image (and optionally creating a shapefile). Select the desired image bands (RGB, MS) based on available bands for computing RGB and/or multispectral indexes. Use the dropdown menu to choose the specific vegetation index, such as NDVI, or create a personalized index using the mosaic layer names. By default, "Compute indexes in memory?" is checked for faster processing. Unchecking this box will allow for index computation using layer-by-layer operations, which may work better for large mosaics (eg., > 4GB) or low RAM availability. Click "Compute the index(es)" to generate the index. The computed indexes can be optionally truncated, eg., to remove soil, and viewed in the various tabs, such as "Plot Index (raster)" and "Plot Index (shapefile)," as shown in the bottom images. Finally, the computed indexes can be downloaded and are available for further analysis. To get access to the app, install the development version of pliman and plimanshiny with install.packages("pak") pak::pkg_install("TiagoOlivoto/pliman") pak::pkg_install("TiagoOlivoto/plimanshiny") Load one of the packages and run library(plimanshiny) run_app() https://lnkd.in/dumN3KyP #plimanshiny #RemoteSensing #RDevelopment #Shiny #DataAnalysis #GIS #DataScience #ExcitingTimes #StayTuned #pliman #digitalfarming #digitalagronomy #remotesensing #uavs #interactive #visualization #rprogramming #spatialanalysis #rstats #phenotyping #precisionagriculture #imageanalysis #mapview #mapedit #labimages #canopy #geospatial #RemoteSensing #GIS #rspatial #buffer #spatialdata #innovation #orthomosaicimages #digitalagriculture #spatialvariability #plantbreeding #dataanalysis #precisionagriculture 🌟 Stay Tuned for More: Exciting updates, tutorials, and the official launch are on the horizon! Follow #plimanshiny for the latest news. If you have a suggestion or criticism to improve the App, don't hesitate to get in touch with me, and let's together improve the pipeline!

100