AI Technology & Industry Review
In a recent technical report, OpenAI introduces Sora, a groundbreaking text-to-video model. Sora stands out for its ability to generate videos and images spanning a wide range of durations, aspect ratios, and resolutions, producing up to a minute of high-definition video content.
In the new paper Unifying Language Learning Paradigms, a Google Research/Brain team proposes a framework for pretraining universal language models that are effective across many different tasks. Their 20B parameter model surpasses 175B GPT-3 on the zero-shot SuperGLUE benchmark and triples the performance of T5-XXL on one-shot summarization tasks.
In the new paper Building Machine Translation Systems for the Next Thousand Languages, a Google Research team proposes a practical machine translation (MT) system that can translate over one thousand languages, including both high-resource and low-resource languages.
In the new paper i-Code: An Integrative and Composable Multimodal Learning Framework, a Microsoft Azure Cognitive Services Research team presents i-Code, a self-supervised pretraining framework that enables the flexible integration of vision, speech, and language modalities and learns their vector representations in a unified manner.
A research team from Rikkyo University and AnyTech Co., Ltd. examines the suitability of different inductive biases for computer vision and proposes Sequencer, an architectural alternative to ViTs that leverages long short-term memory (LSTM) rather than self-attention layers to achieve ViT-competitive performance on long sequence modelling.
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 Flamingo: a Visual Language Model for Few-Shot Learning, a DeepMind research team presents Flamingo, a novel family of visual language models (VLMs) that can handle multimodal tasks such as captioning, visual dialogue, classification and visual question answering when given only a few input/output samples.
Waymo and Google researchers’ new paper A Polynomial Expansion Perspective of Classification Loss Functions presents PolyLoss, a novel and simple framework that redesigns loss functions as a linear combination of polynomial functions that can be tailored to different target tasks and datasets.
In the new paper Expanding the Latent Space of StyleGAN for Real Face Editing, a research team from Northeastern University and Microsoft presents a novel two-branch method that expands the latent space of StyleGAN to enable identity-preserving and disentangled-attribute editing for real face images. The proposed approach achieves both qualitative and quantitative improvements over state-of-the-art methods.
A research team from BIGO Technology and iQIYI Inc. presents ClothFormer, a novel video virtual try-on framework that preserves clothes’ and humans’ features and details to generate realistic and temporally smooth try-on videos that surpass the outputs of current state-of-the-art virtual try-on systems by a large margin.
In the new paper Unified Pretraining Framework for Document Understanding, an Adobe Research and Adobe Document Cloud team presents a unified pretraining framework for document understanding that enables cross-modal connections, relevant information highlighting in both visual and textual modalities, and cross-modal connections. UDoc achieves impressive performance on various downstream tasks.
A research team from the University of Tokyo addresses the challenge of deepfake detection in their new paper Detecting Deepfakes with Self-Blended Images, proposing self-blended images (SBIs), a novel synthetic training data approach that outperforms state-of-the-art methods on unseen manipulations and scenes for deepfake detection tasks.
A DeepMind research team presents a framework for the fine-grained analysis of various distributions shifts and provides insights on when and why we can expect models to successfully generalize.
In the new paper PP-Matting: High-Accuracy Natural Image Matting, a Baidu research team proposes PP-Matting, a trimap-free architecture that combines a high-resolution detail branch and a semantic context branch to achieve state-of-the-art performance on natural image matting tasks.
A research team from DeepMind, Mila – University of Montreal and Google Brain proposes a neural network architecture that learns the graph structure of observational and/or interventional data via supervised training on synthetic graphs, making causal induction a black-box problem that generalizes well to new synthetic and naturalistic graphs.
In the new paper Dancing Under the Stars: Video Denoising in Starlight, a research team from UC Berkeley and Intel Labs leverages a GAN-tuned, physics-based noise model to represent camera noise under low light conditions and trains a novel denoiser that, for the first time, achieves photorealistic video denoising in starlight.
In the new paper A Modern Self-Referential Weight Matrix That Learns to Modify Itself, a research team from The Swiss AI Lab, IDSIA, University of Lugano (USI) & SUPSI, and King Abdullah University of Science and Technology (KAUST) presents a scalable self-referential weight matrix (SRWM) that leverages outer products and the delta update rule to update and improve itself.
In the new paper DeepDPM: Deep Clustering With an Unknown Number of Clusters, a research team from the Ben-Gurion University of the Negev presents DeepDPM, an effective deep nonparametric approach that removes the need to predefine the number of clusters in clustering tasks and can infer it instead.
In the new paper Solving ImageNet: a Unified Scheme for Training any Backbone to Top Results, a research team from Alibaba Group’s DAMO Academy introduces USI (Unified Scheme for ImageNet), a unified scheme for training any backbone on ImageNet that does not require adjustments or hyperparameter tuning between different models, and consistently yields top model results in terms of accuracy and efficiency.
In the new paper Hierarchical Text-Conditional Image Generation with CLIP Latents, an OpenAI research team combines the advantages of contrastive and diffusion models for text-conditional image generation tasks. Their proposed unCLIP model improves image diversity with minimal loss in photorealism and caption similarity, and produces image quality comparable to the state-of-the-art text-to-image system GLIDE.
In the new paper Socratic Models: Composing Zero-Shot Multimodal Reasoning with Language, Google researchers argue that the diversity of different foundation models is symbiotic and that it is possible to build a framework that uses structured Socratic dialogue between pre-existing foundation models to formulate new multimodal tasks as a guided exchange between the models without additional finetuning.
A team from the University of Maryland and Google Research proposes LilNetX, an end-to-end trainable technique for neural networks that jointly optimizes model parameters for accuracy, model size on the disk, and computation on any given task.
The Swiss Federal Institute of Technology Lausanne (EPFL) presents Multi-modal Multi-task Masked Autoencoders (MultiMAE), a simple and effective pretraining strategy that enables masked autoencoding to include multiple modalities and tasks and is applicable to any RGB dataset.
A Meta AI research team explores the plain, non-hierarchical vision transformer (ViT) as a backbone network for object detection, proposing a ViT Detector that achieves performance competitive with traditional hierarchical backbones.
A Google Research team further explores the scaling approach for improving language modelling, leveraging the new Pathways distributed ML system to train a 540 billion parameter autoregressive transformer, Pathways Language Model (PaLM), that achieves state-of-the-art few-shot performance.
Baidu researchers introduce the PP-YOLOE object detector, which outperforms last year’s YOLOX in terms of speed and accuracy trade-off. The PP-YOLOE-l variant surpasses PP-YOLOv2 by 1.9 percent AP and YOLOX-l by 1.3 percent AP on COCO datasets.
In the new paper Training Compute-Optimal Large Language Models, a DeepMind research team posits that current large language models are significantly undertrained and, based on empirical outcomes of over 400 training runs, proposes three predictive approaches for optimally setting model size and training duration.
Researchers from Cash App Labs introduce simple modifications to the Very Deep Variational Autoencoder (VAE) that speedup convergence by 2.6x, save up to 20x in memory, and improve stability during training. Their modified VDVAE achieves state-of-the-art performance on seven commonly used image datasets.
A Stanford research team proposes Time Control (TC), a language model that implicitly plans via a latent stochastic process and generates texts consistent with this latent plan to improve performance on long text generation.
A research team from Carnegie Mellon University and Google systematically explores strategies for leveraging the relatively under-studied resource of bilingual lexicons to adapt pretrained multilingual models to low-resource languages. Their resulting Lexicon-based Adaptation approach produces consistent performance improvements without requiring additional monolingual text.
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.
Researchers from IBM Quantum propose a quantum algorithm for sampling from distributions that can be both complicated and useful, applying the algorithm to perform Markov Chain Monte Carlo (MCMC) iterative sampling on the Boltzmann distribution of classical Ising models.
A Microsoft Research team proposes FocalNet (Focal Modulation Network), a simple and attention-free architecture designed to replace transformers’ self-attention module. FocalNets exhibit significant superiority over self-attention for effective and efficient visual modelling in real-world applications.
A research team from Hotg.ai and Ghent University explores current challenges facing TinyML and techniques to reduce the compute, memory, and energy costs of ML models, providing insights for the large-scale deployment of edge AI.
A Tsinghua University research team proposes Stochastic Scheduled SAM (SS-SAM), a novel and efficient DNN training scheme that achieves comparable or better model training performance with much lower computation cost compared to baseline sharpness-aware minimization (SAM) training schema.
A DeepMind research team argues that the mathematical description of symmetries in group theory is an important foundation that determines the structure of the universe, constrains the nature of natural tasks, and consequently shapes both biological and artificial intelligence. The study proposes symmetry transformations as a fundamental principle for defining what makes good representations.
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.