Search

Time series forecasting requires balancing short-term and long-term dependencies for accurate predictions. Existing methods mainly focus on long-term dependency modeling, neglecting the complexities of short-term dynamics, which may hinder performance. Transformers are superior in modeling long-term dependencies but are criticized for their quadratic computational cost. Mamba provides a near-linear alternative but is reported less effective in time series longterm forecasting due to potential information loss. Current architectures fall short in offering both high efficiency and strong performance for long-term dependency modeling. To address these challenges, we introduce Mixture of Universals (MoU), a versatile model to capture both short-term and long-term dependencies for enhancing performance in time series forecasting. MoU is composed of two novel designs: Mixture of Feature Extractors (MoF), an adaptive method designed to improve time series patch representations for short-term dependency, and Mixture of Architectures (MoA), which hierarchically integrates Mamba, FeedForward, Convolution, and Self-Attention architectures in a specialized order to model long-term dependency from a hybrid perspective. The proposed approach achieves state-of-the-art performance while maintaining relatively low computational costs. Extensive experiments on seven real-world datasets demonstrate the superiority of MoU. Code is available at https://github.com/lunaaa95/mou/., Comment: Code at https://github.com/lunaaa95/mou/

This work presents a Fully BInarized Large Language Model (FBI-LLM), demonstrating for the first time how to train a large-scale binary language model from scratch (not the partial binary or ternary LLM like BitNet b1.58) to match the performance of its full-precision counterparts (e.g., FP16 or BF16) in transformer-based LLMs. It achieves this by employing an autoregressive distillation (AD) loss with maintaining equivalent model dimensions (130M, 1.3B, 7B) and training data volume as regular LLM pretraining, while delivering competitive results in terms of perplexity and task-specific effectiveness. Intriguingly, by analyzing the training trajectory, we find that the pretrained weight is not necessary for training binarized LLMs from scratch. This research encourages a new computational framework and may facilitate the future design of specialized hardware tailored for fully 1-bit LLMs. We make all models, code, and training dataset fully accessible and transparent to support further research (Code: https://github.com/LiqunMa/FBI-LLM. Model: https://huggingface.co/LiqunMa/)., Comment: Github at https://github.com/LiqunMa/FBI-LLM

Multimodal large language models (MLLMs) have shown impressive success across modalities such as image, video, and audio in a variety of understanding and generation tasks. However, current MLLMs are surprisingly poor at understanding webpage screenshots and generating their corresponding HTML code. To address this problem, we propose Web2Code, a benchmark consisting of a new large-scale webpage-to-code dataset for instruction tuning and an evaluation framework for the webpage understanding and HTML code translation abilities of MLLMs. For dataset construction, we leverage pretrained LLMs to enhance existing webpage-to-code datasets as well as generate a diverse pool of new webpages rendered into images. Specifically, the inputs are webpage images and instructions, while the responses are the webpage's HTML code. We further include diverse natural language QA pairs about the webpage content in the responses to enable a more comprehensive understanding of the web content. To evaluate model performance in these tasks, we develop an evaluation framework for testing MLLMs' abilities in webpage understanding and web-to-code generation. Extensive experiments show that our proposed dataset is beneficial not only to our proposed tasks but also in the general visual domain, while previous datasets result in worse performance. We hope our work will contribute to the development of general MLLMs suitable for web-based content generation and task automation. Our data and code will be available at https://github.com/MBZUAI-LLM/web2code., Comment: Website at https://mbzuai-llm.github.io/webpage2code/

Multiple-choice questions (MCQ) are frequently used to assess large language models (LLMs). Typically, an LLM is given a question and selects the answer deemed most probable after adjustments for factors like length. Unfortunately, LLMs may inherently favor certain answer choice IDs, such as A/B/C/D, due to inherent biases of priori unbalanced probabilities, influencing the prediction of answers based on these IDs. Previous research has introduced methods to reduce this ''selection bias'' by simply permutating options on a few test samples and applying to new ones. Another problem of MCQ is the lottery ticket choice by ''random guessing''. The LLM does not learn particular knowledge, but the option is guessed correctly. This situation is especially serious for those small-scale LLMs. To address them, a more thorough approach involves shifting from MCQ to open-style questions, which can fundamentally eliminate selection bias and random guessing issues. However, transitioning causes its own set of challenges in (1) identifying suitable open-style questions and (2) validating the correctness of LLM open-style responses against human-annotated ground-truths. This work aims to tackle these significant difficulties, and establish a new LLM evaluation benchmark through entirely open-style questions. Consequently, we introduce the Open-LLM-Leaderboard to track various LLMs' performance and reflect true capability of them, such as GPT-4o/4/3.5, Claude 3, Gemini, etc. Our code and dataset are available at https://github.com/VILA-Lab/Open-LLM-Leaderboard., Comment: Code and dataset are available at https://github.com/VILA-Lab/Open-LLM-Leaderboard

Dataset condensation, a concept within data-centric learning, efficiently transfers critical attributes from an original dataset to a synthetic version, maintaining both diversity and realism. This approach significantly improves model training efficiency and is adaptable across multiple application areas. Previous methods in dataset condensation have faced challenges: some incur high computational costs which limit scalability to larger datasets (e.g., MTT, DREAM, and TESLA), while others are restricted to less optimal design spaces, which could hinder potential improvements, especially in smaller datasets (e.g., SRe2L, G-VBSM, and RDED). To address these limitations, we propose a comprehensive design framework that includes specific, effective strategies like implementing soft category-aware matching and adjusting the learning rate schedule. These strategies are grounded in empirical evidence and theoretical backing. Our resulting approach, Elucidate Dataset Condensation (EDC), establishes a benchmark for both small and large-scale dataset condensation. In our testing, EDC achieves state-of-the-art accuracy, reaching 48.6% on ImageNet-1k with a ResNet-18 model at an IPC of 10, which corresponds to a compression ratio of 0.78%. This performance exceeds those of SRe2L, G-VBSM, and RDED by margins of 27.3%, 17.2%, and 6.6%, respectively.

Proprietary large language models (LLMs) have been widely applied in various scenarios. Additionally, deploying LLMs on edge devices is trending for efficiency and privacy reasons. However, edge deployment of proprietary LLMs introduces new security challenges: edge-deployed models are exposed as white-box accessible to users, enabling adversaries to conduct effective model stealing (MS) attacks. Unfortunately, existing defense mechanisms fail to provide effective protection. Specifically, we identify four critical protection properties that existing methods fail to simultaneously satisfy: (1) maintaining protection after a model is physically copied; (2) authorizing model access at request level; (3) safeguarding runtime reverse engineering; (4) achieving high security with negligible runtime overhead. To address the above issues, we propose TransLinkGuard, a plug-and-play model protection approach against model stealing on edge devices. The core part of TransLinkGuard is a lightweight authorization module residing in a secure environment, e.g., TEE. The authorization module can freshly authorize each request based on its input. Extensive experiments show that TransLinkGuard achieves the same security protection as the black-box security guarantees with negligible overhead., Comment: arXiv admin note: text overlap with arXiv:2310.07152 by other authors

Dataset distillation aims to compress information from a large-scale original dataset to a new compact dataset while striving to preserve the utmost degree of the original data informational essence. Previous studies have predominantly concentrated on aligning the intermediate statistics between the original and distilled data, such as weight trajectory, features, gradient, BatchNorm, etc. In this work, we consider addressing this task through the new lens of model informativeness in the compression stage on the original dataset pretraining. We observe that with the prior state-of-the-art SRe$^2$L, as model sizes increase, it becomes increasingly challenging for supervised pretrained models to recover learned information during data synthesis, as the channel-wise mean and variance inside the model are flatting and less informative. We further notice that larger variances in BN statistics from self-supervised models enable larger loss signals to update the recovered data by gradients, enjoying more informativeness during synthesis. Building on this observation, we introduce SC-DD, a simple yet effective Self-supervised Compression framework for Dataset Distillation that facilitates diverse information compression and recovery compared to traditional supervised learning schemes, further reaps the potential of large pretrained models with enhanced capabilities. Extensive experiments are conducted on CIFAR-100, Tiny-ImageNet and ImageNet-1K datasets to demonstrate the superiority of our proposed approach. The proposed SC-DD outperforms all previous state-of-the-art supervised dataset distillation methods when employing larger models, such as SRe$^2$L, MTT, TESLA, DC, CAFE, etc., by large margins under the same recovery and post-training budgets. Code is available at https://github.com/VILA-Lab/SRe2L/tree/main/SCDD/.

We present FerKD, a novel efficient knowledge distillation framework that incorporates partial soft-hard label adaptation coupled with a region-calibration mechanism. Our approach stems from the observation and intuition that standard data augmentations, such as RandomResizedCrop, tend to transform inputs into diverse conditions: easy positives, hard positives, or hard negatives. In traditional distillation frameworks, these transformed samples are utilized equally through their predictive probabilities derived from pretrained teacher models. However, merely relying on prediction values from a pretrained teacher, a common practice in prior studies, neglects the reliability of these soft label predictions. To address this, we propose a new scheme that calibrates the less-confident regions to be the context using softened hard groundtruth labels. Our approach involves the processes of hard regions mining + calibration. We demonstrate empirically that this method can dramatically improve the convergence speed and final accuracy. Additionally, we find that a consistent mixing strategy can stabilize the distributions of soft supervision, taking advantage of the soft labels. As a result, we introduce a stabilized SelfMix augmentation that weakens the variation of the mixed images and corresponding soft labels through mixing similar regions within the same image. FerKD is an intuitive and well-designed learning system that eliminates several heuristics and hyperparameters in former FKD solution. More importantly, it achieves remarkable improvement on ImageNet-1K and downstream tasks. For instance, FerKD achieves 81.2% on ImageNet-1K with ResNet-50, outperforming FKD and FunMatch by remarkable margins. Leveraging better pre-trained weights and larger architectures, our finetuned ViT-G14 even achieves 89.9%. Our code is available at https://github.com/szq0214/FKD/tree/main/FerKD., Comment: ICCV 2023. Github: https://github.com/szq0214/FKD/tree/main/FerKD

This paper introduces 26 guiding principles designed to streamline the process of querying and prompting large language models. Our goal is to simplify the underlying concepts of formulating questions for various scales of large language models, examining their abilities, and enhancing user comprehension on the behaviors of different scales of large language models when feeding into different prompts. Extensive experiments are conducted on LLaMA-1/2 (7B, 13B and 70B), GPT-3.5/4 to verify the effectiveness of the proposed principles on instructions and prompts design. We hope that this work can provide a better guide for researchers working on the prompting of large language models. Project page is available at https://github.com/VILA-Lab/ATLAS., Comment: Github at: https://github.com/VILA-Lab/ATLAS

The recent surge in open-source Large Language Models (LLMs), such as LLaMA, Falcon, and Mistral, provides diverse options for AI practitioners and researchers. However, most LLMs have only released partial artifacts, such as the final model weights or inference code, and technical reports increasingly limit their scope to high-level design choices and surface statistics. These choices hinder progress in the field by degrading transparency into the training of LLMs and forcing teams to rediscover many details in the training process. We present LLM360, an initiative to fully open-source LLMs, which advocates for all training code and data, model checkpoints, and intermediate results to be made available to the community. The goal of LLM360 is to support open and collaborative AI research by making the end-to-end LLM training process transparent and reproducible by everyone. As a first step of LLM360, we release two 7B parameter LLMs pre-trained from scratch, Amber and CrystalCoder, including their training code, data, intermediate checkpoints, and analyses (at https://www.llm360.ai). We are committed to continually pushing the boundaries of LLMs through this open-source effort. More large-scale and stronger models are underway and will be released in the future.

The lightweight "local-match-global" matching introduced by SRe2L successfully creates a distilled dataset with comprehensive information on the full 224x224 ImageNet-1k. However, this one-sided approach is limited to a particular backbone, layer, and statistics, which limits the improvement of the generalization of a distilled dataset. We suggest that sufficient and various "local-match-global" matching are more precise and effective than a single one and has the ability to create a distilled dataset with richer information and better generalization. We call this perspective "generalized matching" and propose Generalized Various Backbone and Statistical Matching (G-VBSM) in this work, which aims to create a synthetic dataset with densities, ensuring consistency with the complete dataset across various backbones, layers, and statistics. As experimentally demonstrated, G-VBSM is the first algorithm to obtain strong performance across both small-scale and large-scale datasets. Specifically, G-VBSM achieves a performance of 38.7% on CIFAR-100 with 128-width ConvNet, 47.6% on Tiny-ImageNet with ResNet18, and 31.4% on the full 224x224 ImageNet-1k with ResNet18, under images per class (IPC) 10, 50, and 10, respectively. These results surpass all SOTA methods by margins of 3.9%, 6.5%, and 10.1%, respectively., Comment: Accepted by CVPR2024

Large Language Models (LLMs) with billions of parameters are prime targets for network pruning, removing some model weights without hurting performance. Prior approaches such as magnitude pruning, SparseGPT, and Wanda, either concentrated solely on weights or integrated weights with activations for sparsity. However, they overlooked the informative gradients derived from pretrained LLMs. In this paper, we present a novel sparsity-centric pruning method for pretrained LLMs, termed Gradient-based Language Model Pruner (GBLM-Pruner). GBLM-Pruner leverages the first-order term of the Taylor expansion, operating in a training-free manner by harnessing properly normalized gradients from a few calibration samples to determine the pruning metric, and substantially outperforms competitive counterparts like SparseGPT and Wanda in multiple benchmarks. Intriguingly, by incorporating gradients, unstructured pruning with our method tends to reveal some structural patterns, which mirrors the geometric interdependence inherent in the LLMs' parameter structure. Additionally, GBLM-Pruner functions without any subsequent retraining or weight updates to maintain its simplicity as other counterparts. Extensive evaluations on LLaMA-1 and LLaMA-2 across various benchmarks show that GBLM-Pruner surpasses magnitude pruning, Wanda and SparseGPT by significant margins. We further extend our approach on Vision Transformer. Our code and models are available at https://github.com/VILA-Lab/GBLM-Pruner., Comment: Code and models at https://github.com/VILA-Lab/GBLM-Pruner

This paper aims to understand the impacts of various data combinations (e.g., web text, Wikipedia, GitHub, books) on the pretraining of large language models using SlimPajama. SlimPajama is a rigorously deduplicated, multi-source dataset, which has been refined and further deduplicated to 627B tokens from the extensive 1.2T token RedPajama dataset contributed by Together. We have termed our research as SlimPajama-DC, an empirical analysis designed to uncover fundamental characteristics and best practices associated with employing SlimPajama in the training of large language models. During our research with SlimPajama, two pivotal observations emerged: (1) Global deduplication vs. local deduplication. We analyze and discuss how global (across different sources of datasets) and local (within the single source of dataset) deduplications affect the performance of trained models. (2) Proportions of highly-deduplicated multi-source datasets in the combination. To study this, we construct six configurations on SlimPajama dataset and train individual ones using 1.3B Cerebras-GPT model with Alibi and SwiGLU. Our best configuration outperforms the 1.3B model trained on RedPajama using the same number of training tokens by a significant margin. All our 1.3B models are trained on Cerebras 16$\times$ CS-2 cluster with a total of 80 PFLOP/s in bf16 mixed precision. We further extend our discoveries (such as increasing data diversity is crucial after global deduplication) on a 7B model with large batch-size training. Our SlimPajama-DC models are available at: https://huggingface.co/MBZUAI-LLM/SlimPajama-DC and the separate SlimPajama-DC datasets are available at: https://huggingface.co/datasets/MBZUAI-LLM/SlimPajama-627B-DC., Comment: Technical report. Models at: https://huggingface.co/MBZUAI-LLM/SlimPajama-DC and dataset at: https://huggingface.co/datasets/MBZUAI-LLM/SlimPajama-627B-DC

We introduce Jais and Jais-chat, new state-of-the-art Arabic-centric foundation and instruction-tuned open generative large language models (LLMs). The models are based on the GPT-3 decoder-only architecture and are pretrained on a mixture of Arabic and English texts, including source code in various programming languages. With 13 billion parameters, they demonstrate better knowledge and reasoning capabilities in Arabic than any existing open Arabic and multilingual models by a sizable margin, based on extensive evaluation. Moreover, the models are competitive in English compared to English-centric open models of similar size, despite being trained on much less English data. We provide a detailed description of the training, the tuning, the safety alignment, and the evaluation of the models. We release two open versions of the model -- the foundation Jais model, and an instruction-tuned Jais-chat variant -- with the aim of promoting research on Arabic LLMs. Available at https://huggingface.co/inception-mbzuai/jais-13b-chat, Comment: Arabic-centric, foundation model, large-language model, LLM, generative model, instruction-tuned, Jais, Jais-chat

We propose a unified point cloud video self-supervised learning framework for object-centric and scene-centric data. Previous methods commonly conduct representation learning at the clip or frame level and cannot well capture fine-grained semantics. Instead of contrasting the representations of clips or frames, in this paper, we propose a unified self-supervised framework by conducting contrastive learning at the point level. Moreover, we introduce a new pretext task by achieving semantic alignment of superpoints, which further facilitates the representations to capture semantic cues at multiple scales. In addition, due to the high redundancy in the temporal dimension of dynamic point clouds, directly conducting contrastive learning at the point level usually leads to massive undesired negatives and insufficient modeling of positive representations. To remedy this, we propose a selection strategy to retain proper negatives and make use of high-similarity samples from other instances as positive supplements. Extensive experiments show that our method outperforms supervised counterparts on a wide range of downstream tasks and demonstrates the superior transferability of the learned representations., Comment: Accepted by ICCV 2023

Recently, the community has made tremendous progress in developing effective methods for point cloud video understanding that learn from massive amounts of labeled data. However, annotating point cloud videos is usually notoriously expensive. Moreover, training via one or only a few traditional tasks (e.g., classification) may be insufficient to learn subtle details of the spatio-temporal structure existing in point cloud videos. In this paper, we propose a Masked Spatio-Temporal Structure Prediction (MaST-Pre) method to capture the structure of point cloud videos without human annotations. MaST-Pre is based on spatio-temporal point-tube masking and consists of two self-supervised learning tasks. First, by reconstructing masked point tubes, our method is able to capture the appearance information of point cloud videos. Second, to learn motion, we propose a temporal cardinality difference prediction task that estimates the change in the number of points within a point tube. In this way, MaST-Pre is forced to model the spatial and temporal structure in point cloud videos. Extensive experiments on MSRAction-3D, NTU-RGBD, NvGesture, and SHREC'17 demonstrate the effectiveness of the proposed method., Comment: Accepted by ICCV 2023

Despite the remarkable performance of Vision Transformers (ViTs) in various visual tasks, the expanding computation and model size of ViTs have increased the demand for improved efficiency during training and inference. To address the heavy computation and parameter drawbacks, quantization is frequently studied in the community as a representative model compression technique and has seen extensive use on CNNs. However, due to the unique properties of CNNs and ViTs, the quantization applications on ViTs are still limited and underexplored. In this paper, we identify the difficulty of ViT quantization on its unique variation behaviors, which differ from traditional CNN architectures. The variations indicate the magnitude of the parameter fluctuations and can also measure outlier conditions. Moreover, the variation behaviors reflect the various sensitivities to the quantization of each module. The quantization sensitivity analysis and comparison of ViTs with CNNs help us locate the underlying differences in variations. We also find that the variations in ViTs cause training oscillations, bringing instability during quantization-aware training (QAT). Correspondingly, we solve the variation problem with an efficient knowledge-distillation-based variation-aware quantization method. The multi-crop knowledge distillation scheme can accelerate and stabilize the training and alleviate the variation's influence during QAT. We also proposed a module-dependent quantization scheme and a variation-aware regularization term to suppress the oscillation of weights. On ImageNet-1K, we obtain a 77.66% Top-1 accuracy on the extremely low-bit scenario of 2-bit Swin-T, outperforming the previous state-of-the-art quantized model by 3.35%., Comment: Code is available at https://github.com/HuangOwen/VVTQ

We present a new dataset condensation framework termed Squeeze, Recover and Relabel (SRe$^2$L) that decouples the bilevel optimization of model and synthetic data during training, to handle varying scales of datasets, model architectures and image resolutions for efficient dataset condensation. The proposed method demonstrates flexibility across diverse dataset scales and exhibits multiple advantages in terms of arbitrary resolutions of synthesized images, low training cost and memory consumption with high-resolution synthesis, and the ability to scale up to arbitrary evaluation network architectures. Extensive experiments are conducted on Tiny-ImageNet and full ImageNet-1K datasets. Under 50 IPC, our approach achieves the highest 42.5% and 60.8% validation accuracy on Tiny-ImageNet and ImageNet-1K, outperforming all previous state-of-the-art methods by margins of 14.5% and 32.9%, respectively. Our approach also surpasses MTT in terms of speed by approximately 52$\times$ (ConvNet-4) and 16$\times$ (ResNet-18) faster with less memory consumption of 11.6$\times$ and 6.4$\times$ during data synthesis. Our code and condensed datasets of 50, 200 IPC with 4K recovery budget are available at https://github.com/VILA-Lab/SRe2L., Comment: NeurIPS 2023 spotlight. Code at https://github.com/VILA-Lab/SRe2L

We present Generalized LoRA (GLoRA), an advanced approach for universal parameter-efficient fine-tuning tasks. Enhancing Low-Rank Adaptation (LoRA), GLoRA employs a generalized prompt module to optimize pre-trained model weights and adjust intermediate activations, providing more flexibility and capability across diverse tasks and datasets. Moreover, GLoRA facilitates efficient parameter adaptation by employing a scalable, modular, layer-wise structure search that learns individual adapter of each layer. Originating from a unified mathematical formulation, GLoRA exhibits strong transfer learning, few-shot learning and domain generalization abilities, as it adapts to new tasks through not only weights but also additional dimensions like activations. Comprehensive experiments demonstrate that GLoRA outperforms all previous methods in natural, specialized, and structured vision benchmarks, achieving superior accuracy with fewer parameters and computations. The proposed method on LLaMA-1 and LLaMA-2 also show considerable enhancements compared to the original LoRA in the language domain. Furthermore, our structural re-parameterization design ensures that GLoRA incurs no extra inference cost, rendering it a practical solution for resource-limited applications. Code and models are available at: https://github.com/Arnav0400/ViT-Slim/tree/master/GLoRA., Comment: Technical report. v2: Add LLaMA-1&2 results. Code and models at https://github.com/Arnav0400/ViT-Slim/tree/master/GLoRA

Introduced by Hinton et al. in 2012, dropout has stood the test of time as a regularizer for preventing overfitting in neural networks. In this study, we demonstrate that dropout can also mitigate underfitting when used at the start of training. During the early phase, we find dropout reduces the directional variance of gradients across mini-batches and helps align the mini-batch gradients with the entire dataset's gradient. This helps counteract the stochasticity of SGD and limit the influence of individual batches on model training. Our findings lead us to a solution for improving performance in underfitting models - early dropout: dropout is applied only during the initial phases of training, and turned off afterwards. Models equipped with early dropout achieve lower final training loss compared to their counterparts without dropout. Additionally, we explore a symmetric technique for regularizing overfitting models - late dropout, where dropout is not used in the early iterations and is only activated later in training. Experiments on ImageNet and various vision tasks demonstrate that our methods consistently improve generalization accuracy. Our results encourage more research on understanding regularization in deep learning and our methods can be useful tools for future neural network training, especially in the era of large data. Code is available at https://github.com/facebookresearch/dropout., Comment: ICML 2023

Masked image modeling (MIM) has been recognized as a strong self-supervised pre-training approach in the vision domain. However, the mechanism and properties of the learned representations by such a scheme, as well as how to further enhance the representations are so far not well-explored. In this paper, we aim to explore an interactive Masked Autoencoders (i-MAE) framework to enhance the representation capability from two aspects: (1) employing a two-way image reconstruction and a latent feature reconstruction with distillation loss to learn better features; (2) proposing a semantics-enhanced sampling strategy to boost the learned semantics in MAE. Upon the proposed i-MAE architecture, we can address two critical questions to explore the behaviors of the learned representations in MAE: (1) Whether the separability of latent representations in Masked Autoencoders is helpful for model performance? We study it by forcing the input as a mixture of two images instead of one. (2) Whether we can enhance the representations in the latent feature space by controlling the degree of semantics during sampling on Masked Autoencoders? To this end, we propose a sampling strategy within a mini-batch based on the semantics of training samples to examine this aspect. Extensive experiments are conducted on CIFAR-10/100, Tiny-ImageNet and ImageNet-1K to verify the observations we discovered. Furthermore, in addition to qualitatively analyzing the characteristics of the latent representations, we examine the existence of linear separability and the degree of semantics in the latent space by proposing two evaluation schemes. The surprising and consistent results demonstrate that i-MAE is a superior framework design for understanding MAE frameworks, as well as achieving better representational ability. Code is available at https://github.com/vision-learning-acceleration-lab/i-mae., Comment: Project page: https://zhiqiangshen.com/projects/i-mae/

Recent advances in self-supervised learning have integrated Masked Image Modeling (MIM) and Siamese Networks into a unified framework that leverages the benefits of both techniques. However, several issues remain unaddressed when applying conventional erase-based masking with Siamese ConvNets. These include (I) the inability to drop uninformative masked regions in ConvNets as they process data continuously, resulting in low training efficiency compared to ViT models; and (II) the mismatch between erase-based masking and the contrastive-based objective in Siamese ConvNets, which differs from the MIM approach. In this paper, we propose a filling-based masking strategy called MixMask to prevent information incompleteness caused by the randomly erased regions in an image in the vanilla masking method. Furthermore, we introduce a flexible loss function design that considers the semantic distance change between two different mixed views to adapt the integrated architecture and prevent mismatches between the transformed input and objective in Masked Siamese ConvNets (MSCN). We conducted extensive experiments on various datasets, including CIFAR-100, Tiny-ImageNet, and ImageNet-1K. The results demonstrate that our proposed framework achieves superior accuracy on linear probing, semi-supervised, and supervised finetuning, outperforming the state-of-the-art MSCN by a significant margin. Additionally, we demonstrate the superiority of our approach in object detection and segmentation tasks. Our source code is available at https://github.com/LightnessOfBeing/MixMask., Comment: Technical report. Code is available at https://github.com/LightnessOfBeing/MixMask

In order to deploy deep models in a computationally efficient manner, model quantization approaches have been frequently used. In addition, as new hardware that supports mixed bitwidth arithmetic operations, recent research on mixed precision quantization (MPQ) begins to fully leverage the capacity of representation by searching optimized bitwidths for different layers and modules in a network. However, previous studies mainly search the MPQ strategy in a costly scheme using reinforcement learning, neural architecture search, etc., or simply utilize partial prior knowledge for bitwidth assignment, which might be biased and sub-optimal. In this work, we present a novel Stochastic Differentiable Quantization (SDQ) method that can automatically learn the MPQ strategy in a more flexible and globally-optimized space with smoother gradient approximation. Particularly, Differentiable Bitwidth Parameters (DBPs) are employed as the probability factors in stochastic quantization between adjacent bitwidth choices. After the optimal MPQ strategy is acquired, we further train our network with entropy-aware bin regularization and knowledge distillation. We extensively evaluate our method for several networks on different hardware (GPUs and FPGA) and datasets. SDQ outperforms all state-of-the-art mixed or single precision quantization with a lower bitwidth and is even better than the full-precision counterparts across various ResNet and MobileNet families, demonstrating the effectiveness and superiority of our method., Comment: ICML 2022

This paper explores the feasibility of finding an optimal sub-model from a vision transformer and introduces a pure vision transformer slimming (ViT-Slim) framework. It can search a sub-structure from the original model end-to-end across multiple dimensions, including the input tokens, MHSA and MLP modules with state-of-the-art performance. Our method is based on a learnable and unified $\ell_1$ sparsity constraint with pre-defined factors to reflect the global importance in the continuous searching space of different dimensions. The searching process is highly efficient through a single-shot training scheme. For instance, on DeiT-S, ViT-Slim only takes ~43 GPU hours for the searching process, and the searched structure is flexible with diverse dimensionalities in different modules. Then, a budget threshold is employed according to the requirements of accuracy-FLOPs trade-off on running devices, and a re-training process is performed to obtain the final model. The extensive experiments show that our ViT-Slim can compress up to 40% of parameters and 40% FLOPs on various vision transformers while increasing the accuracy by ~0.6% on ImageNet. We also demonstrate the advantage of our searched models on several downstream datasets. Our code is available at https://github.com/Arnav0400/ViT-Slim., Comment: CVPR 2022. Code is available at https://github.com/Arnav0400/ViT-Slim

This paper aims to explore the feasibility of neural architecture search (NAS) given only a pre-trained model without using any original training data. This is an important circumstance for privacy protection, bias avoidance, etc., in real-world scenarios. To achieve this, we start by synthesizing usable data through recovering the knowledge from a pre-trained deep neural network. Then we use the synthesized data and their predicted soft-labels to guide neural architecture search. We identify that the NAS task requires the synthesized data (we target at image domain here) with enough semantics, diversity, and a minimal domain gap from the natural images. For semantics, we propose recursive label calibration to produce more informative outputs. For diversity, we propose a regional update strategy to generate more diverse and semantically-enriched synthetic data. For minimal domain gap, we use input and feature-level regularization to mimic the original data distribution in latent space. We instantiate our proposed framework with three popular NAS algorithms: DARTS, ProxylessNAS and SPOS. Surprisingly, our results demonstrate that the architectures discovered by searching with our synthetic data achieve accuracy that is comparable to, or even higher than, architectures discovered by searching from the original ones, for the first time, deriving the conclusion that NAS can be done effectively with no need of access to the original or called natural data if the synthesis method is well designed., Comment: ECCV 2022

While Knowledge Distillation (KD) has been recognized as a useful tool in many visual tasks, such as supervised classification and self-supervised representation learning, the main drawback of a vanilla KD framework is its mechanism, which consumes the majority of the computational overhead on forwarding through the giant teacher networks, making the entire learning procedure inefficient and costly. ReLabel, a recently proposed solution, suggests creating a label map for the entire image. During training, it receives the cropped region-level label by RoI aligning on a pre-generated entire label map, allowing for efficient supervision generation without having to pass through the teachers many times. However, as the KD teachers are from conventional multi-crop training, there are various mismatches between the global label-map and region-level label in this technique, resulting in performance deterioration. In this study, we present a Fast Knowledge Distillation (FKD) framework that replicates the distillation training phase and generates soft labels using the multi-crop KD approach, while training faster than ReLabel since no post-processes such as RoI align and softmax operations are used. When conducting multi-crop in the same image for data loading, our FKD is even more efficient than the traditional image classification framework. On ImageNet-1K, we obtain 79.8% with ResNet-50, outperforming ReLabel by ~1.0% while being faster. On the self-supervised learning task, we also show that FKD has an efficiency advantage. Our project page: http://zhiqiangshen.com/projects/FKD/index.html, source code and models are available at: https://github.com/szq0214/FKD., Comment: Our project page: http://zhiqiangshen.com/projects/FKD/index.html, code and models are available at: https://github.com/szq0214/FKD

The nonuniform quantization strategy for compressing neural networks usually achieves better performance than its counterpart, i.e., uniform strategy, due to its superior representational capacity. However, many nonuniform quantization methods overlook the complicated projection process in implementing the nonuniformly quantized weights/activations, which incurs non-negligible time and space overhead in hardware deployment. In this study, we propose Nonuniform-to-Uniform Quantization (N2UQ), a method that can maintain the strong representation ability of nonuniform methods while being hardware-friendly and efficient as the uniform quantization for model inference. We achieve this through learning the flexible in-equidistant input thresholds to better fit the underlying distribution while quantizing these real-valued inputs into equidistant output levels. To train the quantized network with learnable input thresholds, we introduce a generalized straight-through estimator (G-STE) for intractable backward derivative calculation w.r.t. threshold parameters. Additionally, we consider entropy preserving regularization to further reduce information loss in weight quantization. Even under this adverse constraint of imposing uniformly quantized weights and activations, our N2UQ outperforms state-of-the-art nonuniform quantization methods by 0.5~1.7 on ImageNet, demonstrating the contribution of N2UQ design. Code and models are available at: https://github.com/liuzechun/Nonuniform-to-Uniform-Quantization., Comment: CVPR 2022

We present a neat yet effective recursive operation on vision transformers that can improve parameter utilization without involving additional parameters. This is achieved by sharing weights across the depth of transformer networks. The proposed method can obtain a substantial gain (~2%) simply using naive recursive operation, requires no special or sophisticated knowledge for designing principles of networks, and introduces minimal computational overhead to the training procedure. To reduce the additional computation caused by recursive operation while maintaining the superior accuracy, we propose an approximating method through multiple sliced group self-attentions across recursive layers which can reduce the cost consumption by 10~30% with minimal performance loss. We call our model Sliced Recursive Transformer (SReT), a novel and parameter-efficient vision transformer design that is compatible with a broad range of other designs for efficient ViT architectures. Our best model establishes significant improvement on ImageNet-1K over state-of-the-art methods while containing fewer parameters. The proposed weight sharing mechanism by sliced recursion structure allows us to build a transformer with more than 100 or even 1000 shared layers with ease while keeping a compact size (13~15M), to avoid optimization difficulties when the model is too large. The flexible scalability has shown great potential for scaling up models and constructing extremely deep vision transformers. Code is available at https://github.com/szq0214/SReT., Comment: ECCV 2022, 31 pages with Appendix. Code and models are available at https://github.com/szq0214/SReT (v3: update license and fix arxiv timestamp)

In the genome biology research, regulatory genome modeling is an important topic for many regulatory downstream tasks, such as promoter classification, transaction factor binding sites prediction. The core problem is to model how regulatory elements interact with each other and its variability across different cell types. However, current deep learning methods often focus on modeling genome sequences of a fixed set of cell types and do not account for the interaction between multiple regulatory elements, making them only perform well on the cell types in the training set and lack the generalizability required in biological applications. In this work, we propose a simple yet effective approach for pre-training genome data in a multi-modal and self-supervised manner, which we call GeneBERT. Specifically, we simultaneously take the 1d sequence of genome data and a 2d matrix of (transcription factors x regions) as the input, where three pre-training tasks are proposed to improve the robustness and generalizability of our model. We pre-train our model on the ATAC-seq dataset with 17 million genome sequences. We evaluate our GeneBERT on regulatory downstream tasks across different cell types, including promoter classification, transaction factor binding sites prediction, disease risk estimation, and splicing sites prediction. Extensive experiments demonstrate the effectiveness of multi-modal and self-supervised pre-training for large-scale regulatory genomics data.

The best performing Binary Neural Networks (BNNs) are usually attained using Adam optimization and its multi-step training variants. However, to the best of our knowledge, few studies explore the fundamental reasons why Adam is superior to other optimizers like SGD for BNN optimization or provide analytical explanations that support specific training strategies. To address this, in this paper we first investigate the trajectories of gradients and weights in BNNs during the training process. We show the regularization effect of second-order momentum in Adam is crucial to revitalize the weights that are dead due to the activation saturation in BNNs. We find that Adam, through its adaptive learning rate strategy, is better equipped to handle the rugged loss surface of BNNs and reaches a better optimum with higher generalization ability. Furthermore, we inspect the intriguing role of the real-valued weights in binary networks, and reveal the effect of weight decay on the stability and sluggishness of BNN optimization. Through extensive experiments and analysis, we derive a simple training scheme, building on existing Adam-based optimization, which achieves 70.5% top-1 accuracy on the ImageNet dataset using the same architecture as the state-of-the-art ReActNet while achieving 1.1% higher accuracy. Code and models are available at https://github.com/liuzechun/AdamBNN., Comment: ICML 2021. Code and models are available at https://github.com/liuzechun/AdamBNN

This work aims to empirically clarify a recently discovered perspective that label smoothing is incompatible with knowledge distillation. We begin by introducing the motivation behind on how this incompatibility is raised, i.e., label smoothing erases relative information between teacher logits. We provide a novel connection on how label smoothing affects distributions of semantically similar and dissimilar classes. Then we propose a metric to quantitatively measure the degree of erased information in sample's representation. After that, we study its one-sidedness and imperfection of the incompatibility view through massive analyses, visualizations and comprehensive experiments on Image Classification, Binary Networks, and Neural Machine Translation. Finally, we broadly discuss several circumstances wherein label smoothing will indeed lose its effectiveness. Project page: http://zhiqiangshen.com/projects/LS_and_KD/index.html., Comment: ICLR 2021. Project page: http://zhiqiangshen.com/projects/LS_and_KD/index.html

Previous studies dominantly target at self-supervised learning on real-valued networks and have achieved many promising results. However, on the more challenging binary neural networks (BNNs), this task has not yet been fully explored in the community. In this paper, we focus on this more difficult scenario: learning networks where both weights and activations are binary, meanwhile, without any human annotated labels. We observe that the commonly used contrastive objective is not satisfying on BNNs for competitive accuracy, since the backbone network contains relatively limited capacity and representation ability. Hence instead of directly applying existing self-supervised methods, which cause a severe decline in performance, we present a novel guided learning paradigm from real-valued to distill binary networks on the final prediction distribution, to minimize the loss and obtain desirable accuracy. Our proposed method can boost the simple contrastive learning baseline by an absolute gain of 5.5~15% on BNNs. We further reveal that it is difficult for BNNs to recover the similar predictive distributions as real-valued models when training without labels. Thus, how to calibrate them is key to address the degradation in performance. Extensive experiments are conducted on the large-scale ImageNet and downstream datasets. Our method achieves substantial improvement over the simple contrastive learning baseline, and is even comparable to many mainstream supervised BNN methods. Code is available at https://github.com/szq0214/S2-BNN., Comment: CVPR 2021 camera-ready version. Self-supervised binary neural networks using distillation loss (5.5~15% improvement over contrastive baseline). Code is available at https://github.com/szq0214/S2-BNN

The goal of few-shot learning is to learn a classifier that can recognize unseen classes from limited support data with labels. A common practice for this task is to train a model on the base set first and then transfer to novel classes through fine-tuning (Here fine-tuning procedure is defined as transferring knowledge from base to novel data, i.e. learning to transfer in few-shot scenario.) or meta-learning. However, as the base classes have no overlap to the novel set, simply transferring whole knowledge from base data is not an optimal solution since some knowledge in the base model may be biased or even harmful to the novel class. In this paper, we propose to transfer partial knowledge by freezing or fine-tuning particular layer(s) in the base model. Specifically, layers will be imposed different learning rates if they are chosen to be fine-tuned, to control the extent of preserved transferability. To determine which layers to be recast and what values of learning rates for them, we introduce an evolutionary search based method that is efficient to simultaneously locate the target layers and determine their individual learning rates. We conduct extensive experiments on CUB and mini-ImageNet to demonstrate the effectiveness of our proposed method. It achieves the state-of-the-art performance on both meta-learning and non-meta based frameworks. Furthermore, we extend our method to the conventional pre-training + fine-tuning paradigm and obtain consistent improvement., Comment: AAAI 2021. A search based fine-tuning strategy for few-shot learning

We introduce a simple yet effective distillation framework that is able to boost the vanilla ResNet-50 to 80%+ Top-1 accuracy on ImageNet without tricks. We construct such a framework through analyzing the problems in the existing classification system and simplify the base method ensemble knowledge distillation via discriminators by: (1) adopting the similarity loss and discriminator only on the final outputs and (2) using the average of softmax probabilities from all teacher ensembles as the stronger supervision. Intriguingly, three novel perspectives are presented for distillation: (1) weight decay can be weakened or even completely removed since the soft label also has a regularization effect; (2) using a good initialization for students is critical; and (3) one-hot/hard label is not necessary in the distillation process if the weights are well initialized. We show that such a straight-forward framework can achieve state-of-the-art results without involving any commonly-used techniques, such as architecture modification; outside training data beyond ImageNet; autoaug/randaug; cosine learning rate; mixup/cutmix training; label smoothing; etc. Our method obtains 80.67% top-1 accuracy on ImageNet using a single crop-size of 224x224 with vanilla ResNet-50, outperforming the previous state-of-the-arts by a significant margin under the same network structure. Our result can be regarded as a strong baseline using knowledge distillation, and to our best knowledge, this is also the first method that is able to boost vanilla ResNet-50 to surpass 80% on ImageNet without architecture modification or additional training data. On smaller ResNet-18, our distillation framework consistently improves from 69.76% to 73.19%, which shows tremendous practical values in real-world applications. Our code and models are available at: https://github.com/szq0214/MEAL-V2., Comment: 12 pages. Code and trained models are available at: https://github.com/szq0214/MEAL-V2

Unpaired Image-to-image Translation is a new rising and challenging vision problem that aims to learn a mapping between unaligned image pairs in diverse domains. Recent advances in this field like MUNIT and DRIT mainly focus on disentangling content and style/attribute from a given image first, then directly adopting the global style to guide the model to synthesize new domain images. However, this kind of approaches severely incurs contradiction if the target domain images are content-rich with multiple discrepant objects. In this paper, we present a simple yet effective instance-aware image-to-image translation approach (INIT), which employs the fine-grained local (instance) and global styles to the target image spatially. The proposed INIT exhibits three import advantages: (1) the instance-level objective loss can help learn a more accurate reconstruction and incorporate diverse attributes of objects; (2) the styles used for target domain of local/global areas are from corresponding spatial regions in source domain, which intuitively is a more reasonable mapping; (3) the joint training process can benefit both fine and coarse granularity and incorporates instance information to improve the quality of global translation. We also collect a large-scale benchmark for the new instance-level translation task. We observe that our synthetic images can even benefit real-world vision tasks like generic object detection., Comment: Accepted to CVPR 2019. Project page: http://zhiqiangshen.com/projects/INIT/index.html

Transfer learning aims to solve the data sparsity for a target domain by applying information of the source domain. Given a sequence (e.g. a natural language sentence), the transfer learning, usually enabled by recurrent neural network (RNN), represents the sequential information transfer. RNN uses a chain of repeating cells to model the sequence data. However, previous studies of neural network based transfer learning simply represents the whole sentence by a single vector, which is unfeasible for seq2seq and sequence labeling. Meanwhile, such layer-wise transfer learning mechanisms lose the fine-grained cell-level information from the source domain. In this paper, we proposed the aligned recurrent transfer, ART, to achieve cell-level information transfer. ART is under the pre-training framework. Each cell attentively accepts transferred information from a set of positions in the source domain. Therefore, ART learns the cross-domain word collocations in a more flexible way. We conducted extensive experiments on both sequence labeling tasks (POS tagging, NER) and sentence classification (sentiment analysis). ART outperforms the state-of-the-arts over all experiments., Comment: Published at ICLR 2019

Often the best performing deep neural models are ensembles of multiple base-level networks. Unfortunately, the space required to store these many networks, and the time required to execute them at test-time, prohibits their use in applications where test sets are large (e.g., ImageNet). In this paper, we present a method for compressing large, complex trained ensembles into a single network, where knowledge from a variety of trained deep neural networks (DNNs) is distilled and transferred to a single DNN. In order to distill diverse knowledge from different trained (teacher) models, we propose to use adversarial-based learning strategy where we define a block-wise training loss to guide and optimize the predefined student network to recover the knowledge in teacher models, and to promote the discriminator network to distinguish teacher vs. student features simultaneously. The proposed ensemble method (MEAL) of transferring distilled knowledge with adversarial learning exhibits three important advantages: (1) the student network that learns the distilled knowledge with discriminators is optimized better than the original model; (2) fast inference is realized by a single forward pass, while the performance is even better than traditional ensembles from multi-original models; (3) the student network can learn the distilled knowledge from a teacher model that has arbitrary structures. Extensive experiments on CIFAR-10/100, SVHN and ImageNet datasets demonstrate the effectiveness of our MEAL method. On ImageNet, our ResNet-50 based MEAL achieves top-1/5 21.79%/5.99% val error, which outperforms the original model by 2.06%/1.14%. Code and models are available at: https://github.com/AaronHeee/MEAL, Comment: To appear in AAAI 2019. Code and models are available at: https://github.com/AaronHeee/MEAL

The deployment of deep convolutional neural networks (CNNs) in many real world applications is largely hindered by their high computational cost. In this paper, we propose a novel learning scheme for CNNs to simultaneously 1) reduce the model size; 2) decrease the run-time memory footprint; and 3) lower the number of computing operations, without compromising accuracy. This is achieved by enforcing channel-level sparsity in the network in a simple but effective way. Different from many existing approaches, the proposed method directly applies to modern CNN architectures, introduces minimum overhead to the training process, and requires no special software/hardware accelerators for the resulting models. We call our approach network slimming, which takes wide and large networks as input models, but during training insignificant channels are automatically identified and pruned afterwards, yielding thin and compact models with comparable accuracy. We empirically demonstrate the effectiveness of our approach with several state-of-the-art CNN models, including VGGNet, ResNet and DenseNet, on various image classification datasets. For VGGNet, a multi-pass version of network slimming gives a 20x reduction in model size and a 5x reduction in computing operations., Comment: Accepted by ICCV 2017