Towards Data Science’s Post

Learn how to build a neural network in TensorFlow — just follow along Shreya Rao's step-by-step guide (with a bonus section for anyone interested in a PyTorch implementation as well).

TensorLy-Torch is a PyTorch only library that builds on top of TensorLy which provides out-of-the-box tensor layers. You can read more on TensorLy if you haven't used it previously, which is posted in link [1] provided in the comments. TensorLy-Torch Highlights: • Leverage structure in your data: with tensor layers, you can easily leverage the structure in your data through TRL, TCL, Factorized convolutions and more! • Factorized tensors as first class citizens: you can transparently directly create, manipulate and index factorized tensors and regular (dense) pytorch tensors alike! • Built-in tensor layers: all you have to do is import tensorly torch and include the layers we provide directly within your PyTorch models! • Initialization: initializing tensor decompositions can be tricky but its taken care of it all, whether you want to initialize randomly using our Initialization module or from a pretrained layer. • Tensor hooks: you can easily augment your architectures with our built-in Tensor Hooks. Robustify your network with Tensor Dropout and automatically select the rank end-to-end with L1 Regularization! • All the methods available: more methods are being added to make it easy to compare between the performance of various deep tensor based methods!

This article is all you need to get your Ubuntu 22.04 Deep Learning machine ready with the following: Deep Learning rig with Nvidia Cuda toolkit, cuDNN, Conda, PyTorch and TensorFlow: https://lnkd.in/eMcC2BRk

After gaining knowledge about Neural Networks, I thought it was time to apply my skills to a project or competition. I'm glad to share with You that I participated in the Beginner-Friendly "Digit Recognizer"Kaggle competition and I achieved a score of 97% out of 100%✨ This result would give me more determination and dedication to delve into Computer Vision. Here's the link to my Notebook:https://lnkd.in/eY4xFswu

A major challenge in deep learning is not only designing powerful models but also making them accessible and efficient for practical use, especially on devices with limited computing power due to their rapid evolution. As an alternative to the larger and more complex Vision Transformers (ViT), MobileViT is a hybrid, compact, yet robust solution to this challenge. Our latest blog post aims to provide a comprehensive guide to implementing the MobileViT v1 model from scratch using Keras 3, an approach that ensures compatibility across major frameworks like TensorFlow, PyTorch, and Jax. https://buff.ly/3wnOiZm

DRLQ: A Novel Deep Reinforcement Learning

After an afternoon of laziness, trying to get an LLM to write a super basic function for me, I feel inspired to toss a couple of cents into the void. If current generations of LLMs cannot think logically based on all of the data on the internet (i.e. all of the data), maybe a sequence of text isn't the correct representation to reason over? Computer scientists have known this for years - the correct representation over which reasoning can actually work is recursive, and will including a workspace. Maybe it's time to bring classical CS back to the table? There have already been attempts at such architectures - the LLM/ transformer was born as part of a more elaborate architecture - the Neural Turing machine, which borrowed from classical CS concepts. It's time to try again. #AI #ComputationalLogic #TuringMachines #LLMs

A Gold mine dataset for comuter vision is the ImageNet dataset. It consists of about 14 M hand-labelled annotated images which contains over 22,000 day-to-day categories. Every year ImageNet competition is hosted in which the smaller version of this dataset (with 1000 categories) is used with an aim to accurately classify the images. Many winning solutions of the ImageNet Challenge have used state of the art convolutional neural network architectures to beat the best possible accuracy thresholds. I am learning these popular architectures such as VGG16, 19, ResNet, AlexNet etc. In the end, I have explained how to generate image features using pretrained models and use them in machine learning models.

🧠 Introducing SpikingJelly: Revolutionizing Neuromorphic Computing! 🖥 SpikingJelly is a groundbreaking framework designed to accelerate the development and deployment of Spiking Neural Networks (SNNs). Inspired by the brain's efficiency, SpikingJelly introduces a full-stack toolkit for preprocessing neuromorphic datasets, building deep SNNs, optimizing parameters, and deploying on neuromorphic chips. This innovative framework accelerates SNN training by 11×, offering unmatched flexibility and extensibility for custom model acceleration at low costs. The SpikingJelly framework bridges the gap between traditional ANN methods and neuromorphic computing. By leveraging advanced features like multilevel inheritance and semiautomatic code generation, it supports both CPU and GPU simulations with enhanced CUDA acceleration. This approach ensures high performance and ease of use, making SNN research more accessible and efficient for scientists and developers. With SpikingJelly, the potential for creating energy-efficient SNN-based systems is immense. This framework not only enriches the neuromorphic computing ecosystem but also paves the way for the next generation of machine intelligence, harnessing the power of brain-inspired computing for diverse applications. "SpikingJelly: An open-source machine learning infrastructure platform for spike-based intelligence" Wei Fang, Yanqi Chen, Jianhao Ding, Zhaofei Yu, Timothée Masquelier, Ding Chen, Liwei Huang, Huihui Zhou, Guoqi Li, Yonghong Tian Github Repo: https://lnkd.in/grjxYaWX ScienceDirect Article: https://lnkd.in/gWWzCVyb #SpikingNeuralNetworks #NeuromorphicComputing #EnergyEfficientAI #CUDA #AIResearch

a humble implementation for convolutional neural nets on the most famous dataset for it, the MNIST! the dataset is 70k images for numbers (0 to 9) and the model is trying to learn how to recognize them, that's it https://lnkd.in/dJfGJm4C