AI Technology & Industry Review
In the new paper READ: Recurrent Adaptation of Large Transformers, a Meta AI research team proposes REcurrent ADaption (READ), a lightweight and memory-efficient fine-tuning approach that achieves a 56 percent reduction in memory consumption and an 84 percent reduction in GPU use.
In the new paper MegaByte: Predicting Million-Byte Sequences with Multiscale Transformers, a Meta AI research team presents MegaByte, a multiscale decoder architecture that enables million-byte sequence modelling.
In the new paper Unlimiformer: Long-Range Transformers With Unlimited Length Input, a Carnegie Mellon University research team presents a general approach for improving model performance by augmenting pretrained encoder-decoder transformers with an external datastore to permit inputs of unbounded length.
In the new paper ResiDual: Transformer With Dual Residual Connections, a team from Microsoft Research, Microsoft Azure Translation, and Renmin University of China proposes ResiDual, a novel transformer architecture that fuses the connections in post-layer normalization and pre-layer normalization to exploit the benefits of both while also addressing their limitations.
In the new paper Dissecting Recall of Factual Associations in Auto-Regressive Language Models, a team from Google DeepMind, Tel Aviv University and Google Research investigates how factual associations are stored and extracted internally in transformer-based language models and provides insights on how such models’ factual predictions are formed.
In the new paper DETRs Beat YOLOs on Real-Time Object Detection, a Baidu Inc. research team presents Real-Time Detection Transformer (RT-DETR), a real-time end-to-end object detector that leverages a hybrid encoder and novel IoU-aware query selection to address inference speed delay issues. RT-DETR outperforms YOLO object detectors in both accuracy and speed.
In the new paper MathPrompter: Mathematical Reasoning Using Large Language Models, a Microsoft Research team presents MathPrompter, a novel approach that leverages chain-of-thought (CoT) prompting techniques to improve LLM performance on mathematical reasoning problems and increase confidence in their predictions.
In the new paper Exphormer: Sparse Transformers for Graphs, a team from the University of British Columbia, Google Research and the Alberta Machine Intelligence Institute proposes Exphormer, a class of graph transformers with improved scalability and reduced computational complexity that achieves state-of-the-art performance on graph benchmarks.
In the new paper Accelerating Large Language Model Decoding with Speculative Sampling, a DeepMind research team presents SpS (Speculative Sampling), an algorithm that achieves 2–2.5x decoding speedups on a 70 billion parameter Chinchilla language model. The novel approach maintains sample quality and does not require any modifications to model parameters or architecture.
In the new paper Mnemosyne: Learning to Train Transformers with Transformers, a research team from Google and Columbia University presents Mnemosyne Optimizer, a learning-to-learn system for training entire neural network architectures without any task-specific optimizer tuning.
In the new paper Continual Few-Shot Learning Using HyperTransformers, a Google Research team proposes Continual HyperTransformer, which modifies the recently published HyperTransformer few-shot learning method to sequentially update a convolutional neural network’s weights based on the information in a new task without forgetting the knowledge it learned from previous tasks.
In the new paper Tracr: Compiled Transformers as a Laboratory for Interpretability, a research team from ETH Zurich and DeepMind presents Tracr, a compiler that addresses the absence of ground truth explanations in deep neural network models by “compiling” human readable code to the weights of a transformer model.
In the new paper Muse: Text-To-Image Generation via Masked Generative Transformers, a Google Research team introduces Muse, a transformer-based text-to-image synthesis model that leverages masked image modelling to achieve state-of-the-art performance while being significantly faster than diffusion or autoregressive models.
In the new paper Transformer-Based Learned Optimization, a Google Research and Lund University team presents Optimus, an expressive neural network architecture for learned optimization that captures complex dependencies in the parameter space and achieves competitive results on real-world tasks and benchmark optimization problems.
In the new paper Convexifying Transformers: Improving Optimization and Understanding of Transformer Networks, a Stanford University and Google Research team provides a solid theoretical analysis of transformers’ fundamental mechanisms and introduces a novel convex analytic training framework for improving their optimization.
In the new paper Fast DistilBERT on CPUs, researchers from Intel Corporation and Intel Labs propose a pipeline and hardware-aware extreme compression technique for creating and running fast transformer models on CPUs. The approach achieves impressive speed ups and SOTA performance in production environments.
In the new paper Transformers with Multiresolution Attention Heads (currently under double-blind review for ICLR 2023), researchers propose MrsFormer, a novel transformer architecture that uses Multiresolution-head Attention to approximate output sequences and significantly reduces head redundancy without sacrificing accuracy.
In the new paper Wide Attention Is The Way Forward For Transformers, a research team from the University of Cambridge, Imperial College London, and the University of Oxford challenges the commonly held belief that deeper is better for transformer architectures, demonstrating that wider layers result in superior performance on natural language processing tasks.
In the new paper Foundation Transformers, a Microsoft team proposes a method for true general-purpose modelling. Their Foundation Transformer is a single unified transformer that provides guaranteed training stability and can handle diverse tasks and modalities without performance degradation.
In the new paper EcoFormer: Energy-Saving Attention with Linear Complexity, a Monash University research team presents EcoFormer, an attention mechanism with linear complexity that replaces expensive multiply-accumulate operations with simple accumulations and achieves a 73 percent energy footprint reduction on ImageNet.
In the new paper Knowledge Neurons in Pretrained Transformers, a research team from Peking University and Microsoft Research introduces a knowledge attribution method that identifies the neurons that store factual knowledge in pretrained transformers and leverages these neurons to edit factual knowledge in transformers without any fine-tuning.
In the new paper Efficient Training of Language Models to Fill in the Middle, an OpenAI research team shows that causal decoder-based autoregressive (AR) language models can learn to infill texts via a very simple and straightforward transformation to the training data and without any architectural modifications.
In the new paper Is Attention All NeRF Needs?, a research team from the Indian Institute of Technology Madras and the University of Texas at Austin proposes Generalizable NeRF Transformer (GNT), a pure and universal transformer-based architecture for efficient on-the-fly reconstruction of NeRFs. The work demonstrates that a pure attention mechanism suffices for learning a physically-grounded rendering process.
DeepMind researchers present a precise and compact overview of transformer architectures and formal algorithms in the new paper Formal Algorithms for Transformers.
In the new paper VCT: A Video Compression Transformer, a Google Research team presents an elegantly simple but powerful video compression transformer (VCT) that does not require architectural biases and priors and learns totally from data without any hand-crafting. VCT is easy to implement and outperforms conventional video compression approaches.
In the new paper Extreme Compression for Pre-trained Transformers Made Simple and Efficient, a Microsoft research team introduces XTC, a simple yet effective extreme compression pipeline for pretrained transformers that can achieve state-of-the-art results while reducing model size by 50x.
In the new paper A Probabilistic Interpretation of Transformers, ML Collective researcher Alexander Shim provides a probabilistic explanation of transformers’ exponential dot product attention and contrastive learning based on distributions of the exponential family.
In the new technical report OPT: Open Pre-trained Transformer Language Models, Meta AI open-sources OPT, a suite of decoder-only pretrained transformers ranging from 125M to 175B parameters. The release will enable more researchers to work with large-scale language models to drive the field forward.
In the new paper CogView2: Faster and Better Text-to-Image Generation via Hierarchical Transformers, Tsinghua University and the Beijing Academy of Artificial Intelligence researchers pretrain a Cross-Modal general Language Model (CogLM) for text and image token prediction and finetune it for fast super-resolution. The resulting CogView2 hierarchical text-to-image system achieves significant speedups while generating images with better quality at comparable resolutions.
In the new paper Token Dropping for Efficient BERT Pretraining, a research team from Google, New York University, and the University of Maryland proposes a simple but effective “token dropping” technique that significantly reduces the pretraining cost of transformer models such as BERT without hurting performance on downstream fine-tuning tasks.
A Google research team addresses conventional transformers’ resource-heavy training and fine-tuning requirements for learning new knowledge, proposing Memorizing Transformers as a step toward language models that can simply read and memorize new data at inference time for immediate knowledge acquisition.
A team from Google Research and the Swiss AI Lab IDSIA proposes the Block-Recurrent Transformer, a novel long-sequence processing approach that has the same computation time and parameter count costs as a conventional transformer layer but achieves significant perplexity improvements in language modelling tasks over very long sequences.
Researchers from Meta AI and the State University of New York at Buffalo propose sparsely-activated all-MLP architectures (sMLPs) that achieve training efficiency improvements of up to 2x compared to transformer-based mixture-of-experts (MoE) architectures, transformers, and gMLP.
An Idiap Research Institute team proposes a novel multi-layer perceptron (MLP) model, HyperMixer, as a Green AI alternative to transformers. HyperMixer achieves comparable performance with substantially lower costs in terms of processing time, training data and hyperparameter tuning.
A Microsoft research team proposes DeepNorm, a novel normalization function that improves the stability of transformers to enable scaling that is an order of magnitude deeper (more than 1,000 layers) than previous deep transformers.
A research team from Cornell University and Google Brain introduces FLASH, a model family that achieves quality on par with fully augmented transformers while maintaining linear scalability over the context size on modern accelerators.
A team from Facebook AI Research, UC Berkeley and UCLA proposes Online Decision Transformers (ODT), an RL algorithm based on sequence modelling that incorporates offline pretraining and online finetuning in a unified framework and achieves performance competitive with the state-of-the-art models on the D4RL benchmark.
A Google Research team proposes Masked Generative Image Transformer (MaskGIT), a novel image synthesis paradigm that uses a bidirectional transformer decoder. MaskGIT significantly outperforms state-of-the-art transformer models on the ImageNet dataset and accelerates autoregressive decoding by up to 64x.
A Google Research team explores the effects of scaling both input length and model size at the same time with LongT5, a novel transformer architecture that achieves state-of-the-art performance on long-sequence tasks.
In the new paper Self-attention Does Not Need O(n2) Memory, a Google Research team presents novel and simple algorithms for attention and self-attention that require only constant memory and logarithmic memory and reduce the self-attention memory overhead by 59x for inference and by 32x for differentiation at a sequence length of 16384.