Google Research’s Post

It's Research Week here at Maynooth University, so rather than having just one Hamilton seminar this week we have 12! All of which will take place in our lovely seminar room on the 3rd floor of the Eolas Building in the North Campus each day this week at 10am Topics include: Generative AI and bird species detection; DNA computing, and tidal energy generation, amongst many others. Details here: https://lnkd.in/en4e9Sth #hamiltoninstitute #machinelearning #artificialintelligence #datascience #statistics #seminar

We humans resist change but once we accept it why do we normalize it so fast? Like we now see it as totally normal that  if you organize rocks in a specific way and pass electricity between them we can make them think and draw images at this quality??? Just some random thoughts while reading papers on early-fusion multimodal models. The image is from Meta's Chameleon paper, which I think is a little sneak-peak of future Llama models.

Today I earned my "Classify types of space rocks in random photos by using artificial intelligence" badge!🎉 I’m so proud to be celebrating this achievement and hope this inspires you to start your own Microsoft Learn journey!❤ #PythonAI #ML #DataScience #AIwithPython #MachineLearning #DataAnalysis #PythonDeveloper #AIEngineer #DataScientist #ai #machinelearning #artificialintelligence #deeplearning #neuralnetworks #dataanalysis #datamining #microsoftlearn

We are thrilled to announce the publication of our latest research paper in Transportation Research Part B: " Deep hybrid model with satellite imagery: How to combine demand modeling and computer vision for travel behavior analysis?"! This paper develops a synergetic framework, deep hybrid models, for travel behavior analysis to combine low-dimensional numeric data (i.e. sociodemographics and travel attributes) and high-dimensional urban imagery. Our experimental results show that deep hybrid models significantly outperform both classical demand models and deep learning models in predicting aggregate and disaggregate travel behavior, thus demonstrating data complementarity. Deep hybrid models can also generate new satellite images that do not exist in reality and compute the corresponding economic information. We would like to extend our gratitude to all the collaborators: Qingyi Wang, Shenhao Wang, Yunhan Zheng, Hongzhou Lin, Xiaohu Zhang, Jinhua Zhao, and Joan Walker, and we look forward to more groundbreaking collaborations in the future as we continue to advance the research methods in the field of travel behavior analysis. This article is now available on ScienceDirect: https://lnkd.in/e4HryVjX    #ResearchPublication #TravelBehaviorAnalysis #UrbanImagery #HybridModel #TransportationEquity

Hello ML enthusiasts, In this video, we explore a fascinating use case of novel vision transformers for indoor navigation. The content emphasizes the significance of feature engineering and demonstrates how data can be tailored using cutting-edge deep learning methods, specifically the vision transformer. While vision transformers offer advantages over traditional CNNs, it's worth noting their bulkiness and higher computing power requirements. The VITAL framework introduces a pruned vision transformer tailored for edge devices, striking a balance between accuracy and resource efficiency. https://lnkd.in/gt4AWF3q Access the full paper here :

The future is NeuroSymbolic. The inefficiency of linear approximation by brute force is just too overwhelming. Trying to multiply that inefficiency by using a system that only uses the most inefficient information encoding, namely "Human Language", only amplifies that problem. Its like inefficiency on top of inefficiency. This is why intelligence or autonomous systems will get more and more out of reach, the more people try to solve it end-to-end gradient-based. The future is intelligently designed cognitive architectures.

Recently, Professor Astrid Stuckelberger sparked the imagination of many with a bold claim: the existence of 17 dimensions hidden beneath CERN, where interactions with “entities” from other realms might even be possible. Now, while that may sound straight out of science fiction, each theory about extra dimensions needs more than imagination – it demands mountains of data, simulations, and intricate calculations. Behind every hypothesis lies hard work… and an insatiable curiosity. As a data scientist, the thought of analysing data at CERN is as exhilarating as it is complex. Imagine sifting through patterns that not only explain what we know but might open doors to what we don’t. Machine learning and AI aren’t just there to confirm theories; they’re built to spot anomalies, to find that one odd result that might just point us towards extra dimensions. In many ways, it’s like doing science in a cosmic playground. After all, what drives science if not curiosity? Sometimes, data leads us right to the edge of the known and the unknown – and it’s there that we have a chance to shed just a bit more light on this vast, mysterious universe.

We're thrilled to share a groundbreaking piece of research we've come across - an innovative approach to applying deep reinforcement learning in the complex domain of fluid dynamics. This study unveils a system, dubbed "Box o Flows," designed to navigate the intricacies of fluid behaviors in the physical world, a challenge often deemed too complex for accurate simulation. What stands out to us is the system's ability to derive intricate behaviors from straightforward rewards, highlighting the potential of offline learning to maximize data utility. This research resonates deeply with our ethos at Pivot-al, where we're passionate about harnessing the transformative power of data to solve real-world problems. Check out the full study here https://lnkd.in/exRniXSb

As we have spoken about in great detail on MLST over the years, basic NN architectures have severe computational limitations. They are not Turing machines, they can not perform the types of computation a simple calculator can perform. But what if we could have our cake and eat it? The future might just be neurosymbolic! Dr. Petar Veličković is a rising star scientist at DeepMind and is taking his Geometric Deep Learning framework to the next level in collaboration with his collegues. This is just a teaser clip but there will be lots more info coming.