Every so often, we find the most interesting data science links from around the web and collect them in Data Science Briefings, the DataMiningApps newsletter. Subscribe now for free if you want to be the first to get up to speed on interesting resources.
Make-A-Video is a state-of-the-art AI system that generates videos from text.
- DreamFusion: Text-to-3D using 2D Diffusion
“Our approach requires no 3D training data and no modifications to the image diffusion model, demonstrating the effectiveness of pretrained image diffusion models as priors.”
- Mega: Moving Average Equipped Gated Attention
“Mega is a simple, theoretically grounded, single-head gated attention mechanism equipped with (exponential) moving average to incorporate inductive bias of position-aware local dependencies into the position-agnostic attention mechanism.”
- Talk to Books
Browse passages from books using experimental AI.
- AI Streams
Music created by the AI engines.
- AI Hyperopia
“AI hyperopia is the preference for creating flawed solutions to problems beyond the capability of today’s AI, while ignoring the common everyday problems that existing AI could solve.”
- Stable Diffusion in KerasCV
High-performance image generation using Stable Diffusion in KerasCV. Also see this repo.
Monitor your applications and troubleshoot problems in your deployed applications, an open-source alternative to DataDog, New Relic, etc.
- Is causality the missing piece of the AI puzzle?
Qualcomm AI Research explores fundamental research to combine causality with AI.
- Rethinking SGD’s noise
“I will try to show that the instance of SGD used to solve modern ML problems carries rich particularities.”
- Posits, a New Kind of Number, Improves the Math of AI
The first posit-based processor core gave a ten-thousandfold accuracy boost
- Didact AI
The anatomy of an ML-powered stock picking engine
- Productizing Large Language Models
At Replit we have deployed transformer-based language models of all sizes: ~100m parameter models for search and spam, 1-10B models for a code autocomplete product we call GhostWriter, and 100B+ models for features that require a higher reasoning ability. In this post we’ll talk about what we’ve learned about building and hosting large language models.
A Generative Model of High Quality 3D Textured Shapes Learned from Images
- Using GPT-3 to pathfind in random graphs
How did GPT-3 do? The model found a valid path (or correctly reported no result) a little over 60% of the time.
- Stable Diffusion Based Image Compression
Stable Diffusion makes for a very powerful lossy image compression codec.
- Dream Textures
Stable Diffusion built-in to the Blender shader editor.
- Scaling to trillion-parameter model training on AWS
Contiguous parameter management and prefetched activation offloading expand the MiCS tool kit.
- CSCI 601.771: Self-supervised Statistical Models [open course]
“The rise of massive self-supervised (pre-trained) models has transformed various data-driven fields such as natural language processing, computer vision, robotics, and medical imaging. This advanced graduate course aims to provide a holistic view of the issues related to these models.”
- Forecasting with Decision Trees and Random Forests
Random Forests are flexible and powerful when it comes to tabular data. Do they also work for time-series forecasting? Let’s find out.
- On Physics Informed Learning
“One of the most appealing advances in Machine Learning over the past 5 years concerns the development of physics informed neural networks (PINNs). In essence, these efforts have amounted into methods which allow to enforce governing physical and chemical laws into the training of neural networks.”
- K-Means Clustering
An Explorable Explainer By Yi Zhe Ang
- Time series analysis using GANs
The model in this repository combines a stack of dilated causal convolutional and LSTM layers, and while GANs are quite challenging to train, this approach have produced great results on various types of datasets, such as stock market and weather measurements.
- survex: model-agnostic explainability for survival analysis
In this blog, we’d like to cover how model explainability can help make informed choices when working with survival models by showcasing the capabilities of the survex R package.
- Hierarchical Bayesian Neural Networks
Using Blackjax – a library of samplers for JAX that works on CPU as well as GPU.
- CausalNex is a Python library that uses Bayesian Networks to combine machine learning and domain expertise for causal reasoning
You can use CausalNex to uncover structural relationships in your data, learn complex distributions, and observe the effect of potential interventions.
Bayesian regression models using Stan
- Operationalizing Machine Learning: An Interview Study
How does anyone do MLOps, what are the unaddressed challenges, and what are the implications for tool builders?
- Machine Learning Operations
A page full of interesting references and articles.
- labml.ai Annotated PyTorch Paper Implementations
This is a collection of simple PyTorch implementations of neural networks and related algorithms. These implementations are documented with explanations, and the website renders these as side-by-side formatted notes.
- Automatic Differentiation in 26 lines of Python
Inspired by the gist Automatic Differentiation in 38 lines of Haskell. However, unlike that gist, we are doing reverse-mode autodiff here; the method used by Pytorch, TensorFlow, etc.
- Yann LeCun: From Machine Learning to Autonomous Intelligence (video)