📄 1. Papers – DeepMind researchers Brock et al. introduced
Normalizer-Free Networks (NFNets) in their paper "High-Performance Large-Scale Image Recognition Without Normalization" (
arxiv.org). By getting rid of batch normalization they drastically speed-up the model training, achieving remarkable
89.2% Top-1-Accuracy on Imagenet, but
8.7x faster training compared to Efficientnet-B7.
Yannic Kilcher goes into further detail in his one-man paper discussion group (34 min on
youtube.com).
Image: EA / Bioware
💡 2. Use Cases – Game developer BioWare used
AI-powered texture upscaling to remaster their Mass Effect series. This saved them a lot of time when they visually upgraded their 2007 title for current-gen consoles like the PlayStation 5 (
venturebeat.com).
💻 3. Libraries & Code – Lightning
Flash is a high-level framework for fast prototyping of Deep Learning tasks, built on PyTorch Lightning. It aims to simplify fine-tuning and deployment of state-of-the-art models on custom tasks. Find the release article on Medium (
medium.com/pytorch) and code available on GitHub (
github.com).
🎨 4. Showcase – Jina released version 1.0 of their
open-source Neural Search Engine for enterprises and developers. Their
multi-modal search framework can easily find text, graphics, audio, or video in large-scale cloud operations. Try it yourself with code available on (
jina.ai &
github.com).
💭 4.2 Articles & Tutorials – Or start from scratch and build you own AI-assisted search engine with Elasticsearch, Kubeflow and Katib following this Topbots Tutorial (
topbots.com).
5. 👁 Miscellaneous – To train Facial Recognition Systems one needs large amounts of input images. Some model trainers took the freedom to scrape images from websites like Flickr - and
your picture might be one of them. With
Exposing.ai two tech artists built a platform for you to check whether your face was used to train common face recognisers (
exposing.ai).
🛠 6. Tools – Papers with Code just became Papers + Code + Datasets. The beloved database for code implementation, benchmarks and academic papers, now also features 3000+ research datasets, sorted by task and modality and usage statistics (
paperswithcode.com/datasets).
💡 7. Use Cases – Researchers from Tartu, Estonia trained a GAN to create “artificial genomes”. Instead of creating DeepFakes of human faces, they fake
DNA sequences that are 100%
synthetic but indistinguishable from human genomes. This could relieve institutions of privacy measures they currently have to ensure with real DNA data (
thenextweb.com).
🎨 8. Showcase – Automated Paper Reviews with AI: Waiting for a scientific paper review? Wait no longer. This system can judge your or anybody else's paper automatically (
review.nlpedia.ai).
Image: Le et al.
9. Papers – Dutch researchers Le et al. have attempted to reconstruct Dr. Who scenes from viewers' fMRI brain scans, in "Brain2Pix: Fully convolutional naturalistic video reconstruction from brain activity" (
biorxiv.org).
IN CASE YOU MISSED IT
📍 The
Enterprise AI Consulting team from AMAI launched a new
website. An updated careers page is accompanied by a newly started AI Expert developer blog.