AI Technology & Industry Review
In the new paper Shaking the Foundations: Delusions in Sequence Models for Interaction and Control, a DeepMind research team explores the origins of mismatch problems in sequence models that lack understanding of the cause and effect of their actions, and addresses the problem by treating actions as causal interventions.
Researchers from Fudan University, University of Surrey and Huawei Noah’s Ark Lab identify the limitations of quadratic complexity for vision transformers (ViTs) as rooted in keeping the softmax self-attention during approximations. The team proposes the first softmax-free transformer (SOFT), which reduces the self-attention computation to linear complexity, achieving a superior trade-off between accuracy and complexity.
A research team from Google Brain and Google Research introduces SCENIC, an open-source JAX library for fast and extensible computer vision research and beyond. JAX currently supports implementations of state-of-the-art vision models such as ViT, DETR and MLP Mixer, and more open-sourced cutting-edge projects will be added in the near future.
Facebook AI Research proposes NormFormer, an approach that improves pretraining perplexity and downstream task performance for both causal and masked language models, achieving GPT3-Large (1.3B) zero-shot performance 60 percent faster and improving fine-tuned GLUE performance by 1.9 percent.
A Facebook AI research team releases SaLinA, a reinforcement learning (RL) library for model-based RL, differentiable environments and multi-agent RL that simplifies the implementation of complex sequential learning models.
In the new paper Non-deep Networks, a research team from Princeton University and Intel Labs argues it is possible to achieve high performance with “non-deep” neural networks, presenting ParNet (Parallel Networks), a novel 12-layer architecture that achieves performance competitive with its state-of-the-art deep counterparts.
A DeepMind research team explores the problem of how to train agents to collaborate well with novel human partners without using human data and presents Fictitious Co-Play (FCP), a surprisingly simple approach designed to address this challenge.
Facebook AI and NYU researchers challenge the conventional wisdom regarding interpolation in machine learning, arguing that interpolation almost never happens on high-dimensional datasets.
In a paper currently under double-blind review for ICLR 2022, researchers propose StyleNeRF, a 3D-aware generative model that can synthesize high-resolution images at interactive rates while preserving high-quality 3D consistency, and can even generalize to unseen views with control on styles and poses.
A research team from the University of Southern California and Google proposes TOME, a “mention memory” approach to factual knowledge extraction for NLU tasks. A transformer model with attention over a semi-parametric representation of the entire Wikipedia text corpus, TOME can extract information without supervision and achieves strong performance on multiple open-domain question answering benchmarks.
A Google Research team introduces Autoregressive Diffusion Models (ARDMs), a model class encompassing and generalizing order-agnostic autoregressive models and discrete diffusion models that can generate variables in an arbitrary order and upscale variables.
A Google Research team conducts a systematic exploration comprising more than 4800 experiments on Vision Transformers, MLP-Mixers and ResNets with parameters ranging from 10 million to 10 billion, evaluated on more than 20 downstream image recognition tasks, aiming to capture the nonlinear relationships between performance on upstream and downstream tasks.
On October 13, ICCV 2021 announced its Best Paper Awards, honourable mentions, and Best Student Paper.
A NVIDIA and Aalto University research team presents StyleGAN3, a novel generative adversarial network (GAN) architecture where the exact sub-pixel position of each feature is exclusively inherited from the underlying coarse features, enabling a more natural transformation hierarchy and advancing GAN-based animation generation.
A research team proposes ConvMixer, an extremely simple model designed to support the argument that the impressive performance of vision transformers (ViTs) is mainly attributable to their use of patches as the input representation. The study shows that ConvMixer can outperform ViTs, MLP-Mixers and classical vision models.
On October 11, Microsoft introduced the largest and “the most powerful monolithic transformer language model” trained to date, a 530 billion parameter GPT-3-style generative language model.
According to Reuters, the Pentagon’s former software chief Nicolas Chaillan told the Financial Times on Sunday that “We have no competing fighting chance against China in 15 to 20 years. Right now, it’s already a done deal; it is already over, in my opinion.”
An Apple research team performs a comparative analysis on a contrastive self-supervised learning (SSL) algorithm (SimCLR) and a supervised learning (SL) approach for simple image data in a common architecture, shedding light on the similarities and dissimilarities in their learned visual representation patterns.
A research team from Facebook AI and Carnegie Mellon University presents VideoCLIP, a contrastive approach for pretraining a unified model for zero-shot video and text understanding without requiring annotated data for downstream tasks.
On October 6, Sony Group announced that the company will launch its edge AI platform service Aitrios AI sensing platform in Japan, the U.S. and Europe starting late this year.
A Google Research team presents compressive variants of SimCLR and BYOL that yield better and more robust visual representations.
The London-based AI research firm DeepMind has introduced AlphaFold-Multimer, a model that can predict the structure of multi-chain protein complexes with increased accuracy.
A DeepMind research team proposes Faster Improvement Rate PBT (FIRE PBT) for Population Based Training (PBT), an automated hyperparameter tuning method for neural network training. The novel approach achieves faster improvement rates and better long-term performance.
An Oxford University research team presents PASS, a large (1.28M) image collection excluding humans, created as an ImageNet replacement for self-supervised pretraining without technical, ethical or legal issues.
A research team from ETH Zurich and NVIDIA proposes a training framework that enables fast policy generation for real-world robotic tasks — training time can be reduced by multiple orders of magnitude using massive parallelism on a single workstation GPU.