Medical image segmentation plays a crucial role in advancing healthcare systems for disease diagnosis and treatment planning. The u-shaped architecture, popularly known as U-Net, has proven highly successful for various medical image segmentation tasks. However, U-Net's convolution-based operations inherently limit its ability to model long-range dependencies effectively. To address these limitations, researchers have turned to Transformers, renowned for their global self-attention mechanisms, as alternative architectures. One popular network is our previous TransUNet, which leverages Transformers' self-attention to complement U-Net's localized information with the global context. In this paper, we extend the 2D TransUNet architecture to a 3D network by building upon the state-of-the-art nnU-Net architecture, and fully exploring Transformers' potential in both the encoder and decoder design. We introduce two key components: 1) A Transformer encoder that tokenizes image patches from a convolution neural network (CNN) feature map, enabling the extraction of global contexts, and 2) A Transformer decoder that adaptively refines candidate regions by utilizing cross-attention between candidate proposals and U-Net features. Our investigations reveal that different medical tasks benefit from distinct architectural designs. The Transformer encoder excels in multi-organ segmentation, where the relationship among organs is crucial. On the other hand, the Transformer decoder proves more beneficial for dealing with small and challenging segmented targets such as tumor segmentation. Extensive experiments showcase the significant potential of integrating a Transformer-based encoder and decoder into the u-shaped medical image segmentation architecture. TransUNet outperforms competitors in various medical applications.
Significant progress has been made recently in point cloud segmentation utilizing an encoder-decoder framework, which initially encodes point clouds into low-resolution representations and subsequently decodes high-resolution predictions. Inspired by the success of high-resolution architectures in image dense prediction, which always maintains a high-resolution representation throughout the entire learning process, we consider it also highly important for 3D dense point cloud analysis. Therefore, in this paper, we explore high-resolution architectures for 3D point cloud segmentation. Specifically, we generalize high-resolution architectures using a unified pipeline named PointHR, which includes a knn-based sequence operator for feature extraction and a differential resampling operator to efficiently communicate different resolutions. Additionally, we propose to avoid numerous on-the-fly computations of high-resolution architectures by pre-computing the indices for both sequence and resampling operators. By doing so, we deliver highly competitive high-resolution architectures while capitalizing on the benefits of well-designed point cloud blocks without additional effort. To evaluate these architectures for dense point cloud analysis, we conduct thorough experiments using S3DIS and ScanNetV2 datasets, where the proposed PointHR outperforms recent state-of-the-art methods without any bells and whistles. The source code is available at \url{https://github.com/haibo-qiu/PointHR}.
Over the past years, foundation models have caused a paradigm shift in machine learning due to their unprecedented capabilities for zero-shot and few-shot generalization. However, despite the success of foundation models in modalities such as natural language processing and computer vision, the development of foundation models for time series forecasting has lagged behind. We present Lag-Llama, a general-purpose foundation model for univariate probabilistic time series forecasting based on a decoder-only transformer architecture that uses lags as covariates. Lag-Llama is pretrained on a large corpus of diverse time series data from several domains, and demonstrates strong zero-shot generalization capabilities compared to a wide range of forecasting models on downstream datasets across domains. Moreover, when fine-tuned on relatively small fractions of such previously unseen datasets, Lag-Llama achieves state-of-the-art performance, outperforming prior deep learning approaches, emerging as the best general-purpose model on average. Lag-Llama serves as a strong contender to the current state-of-art in time series forecasting and paves the way for future advancements in foundation models tailored to time series data.
We study the problem of building a controller that can follow open-ended instructions in open-world environments. We propose to follow reference videos as instructions, which offer expressive goal specifications while eliminating the need for expensive text-gameplay annotations. A new learning framework is derived to allow learning such instruction-following controllers from gameplay videos while producing a video instruction encoder that induces a structured goal space. We implement our agent GROOT in a simple yet effective encoder-decoder architecture based on causal transformers. We evaluate GROOT against open-world counterparts and human players on a proposed Minecraft SkillForge benchmark. The Elo ratings clearly show that GROOT is closing the human-machine gap as well as exhibiting a 70% winning rate over the best generalist agent baseline. Qualitative analysis of the induced goal space further demonstrates some interesting emergent properties, including the goal composition and complex gameplay behavior synthesis. The project page is available at https://craftjarvis-groot.github.io.
Defect detection is a critical research area in artificial intelligence. Recently, synthetic data-based self-supervised learning has shown great potential on this task. Although many sophisticated synthesizing strategies exist, little research has been done to investigate the robustness of models when faced with different strategies. In this paper, we focus on this issue and find that existing methods are highly sensitive to them. To alleviate this issue, we present a Discrepancy Aware Framework (DAF), which demonstrates robust performance consistently with simple and cheap strategies across different anomaly detection benchmarks. We hypothesize that the high sensitivity to synthetic data of existing self-supervised methods arises from their heavy reliance on the visual appearance of synthetic data during decoding. In contrast, our method leverages an appearance-agnostic cue to guide the decoder in identifying defects, thereby alleviating its reliance on synthetic appearance. To this end, inspired by existing knowledge distillation methods, we employ a teacher-student network, which is trained based on synthesized outliers, to compute the discrepancy map as the cue. Extensive experiments on two challenging datasets prove the robustness of our method. Under the simple synthesis strategies, it outperforms existing methods by a large margin. Furthermore, it also achieves the state-of-the-art localization performance. Code is available at: https://github.com/caiyuxuan1120/DAF.
Image captioning is a challenging task involving generating a textual description for an image using computer vision and natural language processing techniques. This paper proposes a deep neural framework for image caption generation using a GRU-based attention mechanism. Our approach employs multiple pre-trained convolutional neural networks as the encoder to extract features from the image and a GRU-based language model as the decoder to generate descriptive sentences. To improve performance, we integrate the Bahdanau attention model with the GRU decoder to enable learning to focus on specific image parts. We evaluate our approach using the MSCOCO and Flickr30k datasets and show that it achieves competitive scores compared to state-of-the-art methods. Our proposed framework can bridge the gap between computer vision and natural language and can be extended to specific domains.
Pretraining auto-regressive large language models~(LLMs) with retrieval demonstrates better perplexity and factual accuracy by leveraging external databases. However, the size of existing pretrained retrieval-augmented LLM is still limited (e.g., Retro has 7.5B parameters), which limits the effectiveness of instruction tuning and zero-shot generalization. In this work, we introduce Retro 48B, the largest LLM pretrained with retrieval. Specifically, we continue to pretrain a 43B GPT model on additional 100 billion tokens using the Retro augmentation method by retrieving from 1.2 trillion tokens. Notably, the obtained foundation model, Retro 48B, largely outperforms the counterpart GPT 43B trained on 1.2T tokens in terms of perplexity with only 2.58% additional GPU hours, demonstrating the significant scaling potential of the method. After instruction tuning on Retro, InstructRetro demonstrates significant improvement over the instruction tuned GPT on a wide range of zero-shot tasks. Specifically, the average improvement of InstructRetro is 7% over its GPT counterpart across 8 short-form QA and reading comprehension tasks, 10% over GPT across 4 challenging long-form QA tasks, and 16% over GPT across 3 summarization tasks. Surprisingly, we find that one can ablate the encoder from InstructRetro architecture and directly use its decoder backbone, while achieving comparable results. Our results highlight the promising direction to obtain a better GPT decoder through continued pretraining with retrieval before instruction tuning. Our code and checkpoints are publicly available at: https://huggingface.co/nvidia/retro-48b-instruct-4k.
Foundation models are applied in a broad spectrum of settings with different inference constraints, from massive multi-accelerator clusters to resource-constrained standalone mobile devices. However, the substantial costs associated with training these models often limit the number of unique model sizes that can be offered. Consequently, practitioners are compelled to select a model that may not be optimally aligned with their specific latency and cost requirements. We present MatFormer, a novel Transformer architecture designed to provide elastic inference across diverse deployment constraints. MatFormer achieves this by incorporating a nested Feed Forward Network (FFN) block structure within a standard Transformer model. During training, we optimize the parameters of multiple nested FFN blocks with varying sizes, enabling the extraction of hundreds of accurate smaller models without incurring additional computational costs. We empirically validate the efficacy of MatFormer across different model classes (decoders and encoders) and modalities (language and vision), demonstrating its potential for real-world deployment. We show that a 850M decoder-only MatFormer language model (MatLM) allows us to extract multiple smaller models spanning from 582M to 850M parameters, each exhibiting better validation loss and one-shot downstream evaluations than independently trained counterparts. Furthermore, we observe that smaller encoders extracted from a universal MatFormer-based ViT (MatViT) encoder preserve the metric-space structure for adaptive large-scale retrieval. Finally, we showcase that speculative decoding with the accurate and consistent submodels extracted from MatFormer can lead to significant reduction in inference latency. Project website: https://devvrit.github.io/matformer/
Deep learning based decoding networks have shown significant improvement in decoding LDPC codes, but the neural decoders are limited by rate-matching operations such as puncturing or extending, thus needing to train multiple decoders with different code rates for a variety of channel conditions. In this correspondence, we propose a Multi-Task Learning based rate-compatible LDPC ecoding network, which utilizes the structure of raptor-like LDPC codes and can deal with multiple code rates. In the proposed network, different portions of parameters are activated to deal with distinct code rates, which leads to parameter sharing among tasks. Numerical experiments demonstrate the effectiveness of the proposed method. Training the specially designed network under multiple code rates makes the decoder compatible with multiple code rates without sacrificing frame error rate performance.
The high efficiency video coding (HEVC) standard and the joint exploration model (JEM) codec incorporate 35 and 67 intra prediction modes (IPMs) respectively, which are essential for efficient compression of Intra coded blocks. These IPMs are transmitted to the decoder through a coding scheme. In our paper, we present an innovative approach to construct a dedicated coding scheme for IPM based on contextual information. This approach comprises three key steps: prediction, clustering, and coding, each of which has been enhanced by introducing new elements, namely, labels for prediction, tests for clustering, and codes for coding. In this context, we have proposed a method that utilizes a genetic algorithm to minimize the rate cost, aiming to derive the most efficient coding scheme while leveraging the available labels, tests, and codes. The resulting coding scheme, expressed as a binary tree, achieves the highest coding efficiency for a given level of complexity. In our experimental evaluation under the HEVC standard, we observed significant bitrate gains while maintaining coding efficiency under the JEM codec. These results demonstrate the potential of our approach to improve compression efficiency, particularly under the HEVC standard, while preserving the coding efficiency of the JEM codec.
Neural shape representation generally refers to representing 3D geometry using neural networks, e.g., computing a signed distance or occupancy value at a specific spatial position. In this paper we present a neural-network architecture suitable for accurate encoding of 3D shapes in a single forward pass. Our architecture is based on a multi-scale hybrid system incorporating graph-based and voxel-based components, as well as a continuously differentiable decoder. The hybrid system includes a novel way of voxelizing point-based features in neural networks, which we show can be used in combination with oriented point-clouds to obtain smoother and more detailed reconstructions. Furthermore, our network is trained to solve the eikonal equation and only requires knowledge of the zero-level set for training and inference. This means that in contrast to most previous shape encoder architectures, our network is able to output valid signed distance fields without explicit prior knowledge of non-zero distance values or shape occupancy. It also requires only a single forward-pass, instead of the latent-code optimization used in auto-decoder methods. We further propose a modification to the loss function in case that surface normals are not well defined, e.g., in the context of non-watertight surfaces and non-manifold geometry, resulting in an unsigned distance field. Overall, our system can help to reduce the computational overhead of training and evaluating neural distance fields, as well as enabling the application to difficult geometry.
Reconstructing 3D human shapes from 2D images has received increasing attention recently due to its fundamental support for many high-level 3D applications. Compared with natural images, freehand sketches are much more flexible to depict various shapes, providing a high potential and valuable way for 3D human reconstruction. However, such a task is highly challenging. The sparse abstract characteristics of sketches add severe difficulties, such as arbitrariness, inaccuracy, and lacking image details, to the already badly ill-posed problem of 2D-to-3D reconstruction. Although current methods have achieved great success in reconstructing 3D human bodies from a single-view image, they do not work well on freehand sketches. In this paper, we propose a novel sketch-driven multi-faceted decoder network termed SketchBodyNet to address this task. Specifically, the network consists of a backbone and three separate attention decoder branches, where a multi-head self-attention module is exploited in each decoder to obtain enhanced features, followed by a multi-layer perceptron. The multi-faceted decoders aim to predict the camera, shape, and pose parameters, respectively, which are then associated with the SMPL model to reconstruct the corresponding 3D human mesh. In learning, existing 3D meshes are projected via the camera parameters into 2D synthetic sketches with joints, which are combined with the freehand sketches to optimize the model. To verify our method, we collect a large-scale dataset of about 26k freehand sketches and their corresponding 3D meshes containing various poses of human bodies from 14 different angles. Extensive experimental results demonstrate our SketchBodyNet achieves superior performance in reconstructing 3D human meshes from freehand sketches.
In this paper, we present our solution to a Multi-modal Algorithmic Reasoning Task: SMART-101 Challenge. Different from the traditional visual question-answering datasets, this challenge evaluates the abstraction, deduction, and generalization abilities of neural networks in solving visuolinguistic puzzles designed specifically for children in the 6-8 age group. We employed a divide-and-conquer approach. At the data level, inspired by the challenge paper, we categorized the whole questions into eight types and utilized the llama-2-chat model to directly generate the type for each question in a zero-shot manner. Additionally, we trained a yolov7 model on the icon45 dataset for object detection and combined it with the OCR method to recognize and locate objects and text within the images. At the model level, we utilized the BLIP-2 model and added eight adapters to the image encoder VIT-G to adaptively extract visual features for different question types. We fed the pre-constructed question templates as input and generated answers using the flan-t5-xxl decoder. Under the puzzle splits configuration, we achieved an accuracy score of 26.5 on the validation set and 24.30 on the private test set.
AI accelerator processing capabilities and memory constraints largely dictate the scale in which machine learning workloads (e.g., training and inference) can be executed within a desirable time frame. Training a state of the art, transformer-based model today requires use of GPU-accelerated high performance computers with high-speed interconnects. As datasets and models continue to increase in size, computational requirements and memory demands for AI also continue to grow. These challenges have inspired the development of distributed algorithm and circuit-based optimization techniques that enable the ability to progressively scale models in multi-node environments, efficiently minimize neural network cost functions for faster convergence, and store more parameters into a set number of available resources. In our research project, we focus on parallel and distributed machine learning algorithm development, specifically for optimizing the data processing and pre-training of a set of 5 encoder-decoder LLMs, ranging from 580 million parameters to 13 billion parameters. We performed a fine-grained study to quantify the relationships between three ML parallelism methods, specifically exploring Microsoft DeepSpeed Zero Redundancy Optimizer (ZeRO) stages.
Learning representations that generalize to novel compositions of known concepts is crucial for bridging the gap between human and machine perception. One prominent effort is learning object-centric representations, which are widely conjectured to enable compositional generalization. Yet, it remains unclear when this conjecture will be true, as a principled theoretical or empirical understanding of compositional generalization is lacking. In this work, we investigate when compositional generalization is guaranteed for object-centric representations through the lens of identifiability theory. We show that autoencoders that satisfy structural assumptions on the decoder and enforce encoder-decoder consistency will learn object-centric representations that provably generalize compositionally. We validate our theoretical result and highlight the practical relevance of our assumptions through experiments on synthetic image data.
Active learning (AL), which aims to construct an effective training set by iteratively curating the most formative unlabeled data for annotation, has been widely used in low-resource tasks. Most active learning techniques in classification rely on the model's uncertainty or disagreement to choose unlabeled data, suffering from the problem of over-confidence in superficial patterns and a lack of exploration. Inspired by the cognitive processes in which humans deduce and predict through causal information, we take an initial attempt towards integrating rationales into AL and propose a novel Explainable Active Learning framework (XAL) for low-resource text classification, which aims to encourage classifiers to justify their inferences and delve into unlabeled data for which they cannot provide reasonable explanations. Specifically, besides using a pre-trained bi-directional encoder for classification, we employ a pre-trained uni-directional decoder to generate and score the explanation. We further facilitate the alignment of the model with human reasoning preference through a proposed ranking loss. During the selection of unlabeled data, the predicted uncertainty of the encoder and the explanation score of the decoder complement each other as the final metric to acquire informative data. Extensive experiments on six datasets show that XAL achieves consistent improvement over 9 strong baselines. Analysis indicates that the proposed method can generate corresponding explanations for its predictions.
Computed tomography (CT) serves as an effective tool for lung cancer screening, diagnosis, treatment, and prognosis, providing a rich source of features to quantify temporal and spatial tumor changes. Nonetheless, the diversity of CT scanners and customized acquisition protocols can introduce significant inconsistencies in texture features, even when assessing the same patient. This variability poses a fundamental challenge for subsequent research that relies on consistent image features. Existing CT image standardization models predominantly utilize GAN-based supervised or semi-supervised learning, but their performance remains limited. We present DiffusionCT, an innovative score-based DDPM model that operates in the latent space to transform disparate non-standard distributions into a standardized form. The architecture comprises a U-Net-based encoder-decoder, augmented by a DDPM model integrated at the bottleneck position. First, the encoder-decoder is trained independently, without embedding DDPM, to capture the latent representation of the input data. Second, the latent DDPM model is trained while keeping the encoder-decoder parameters fixed. Finally, the decoder uses the transformed latent representation to generate a standardized CT image, providing a more consistent basis for downstream analysis. Empirical tests on patient CT images indicate notable improvements in image standardization using DiffusionCT. Additionally, the model significantly reduces image noise in SPAD images, further validating the effectiveness of DiffusionCT for advanced imaging tasks.
Information Extraction (IE) is an essential task in Natural Language Processing. Traditional methods have relied on coarse-grained extraction with simple instructions. However, with the emergence of Large Language Models (LLMs), there is a need to adapt IE techniques to leverage the capabilities of these models. This paper introduces a fine-grained IE benchmark dataset tailored for LLMs, employing augmented instructions for each information type, which includes task descriptions, extraction rules, output formats, and examples. Through extensive evaluations, we observe that encoder-decoder models, particularly T5 and FLAN-T5, perform well in generalizing to unseen information types, while ChatGPT exhibits greater adaptability to new task forms. Our results also indicate that performance is not solely dictated by model scale, and highlight the significance of architecture, data diversity, and learning techniques. This work paves the way for a more refined and versatile utilization of LLMs in Information Extraction.
Semantic segmentation tasks naturally require high-resolution information for pixel-wise segmentation and global context information for class prediction. While existing vision transformers demonstrate promising performance, they often utilize high-resolution context modeling, resulting in a computational bottleneck. In this work, we challenge conventional wisdom and introduce the Low-Resolution Self-Attention (LRSA) mechanism to capture global context at a significantly reduced computational cost, i.e., FLOPs. Our approach involves computing self-attention in a fixed low-resolution space regardless of the input image's resolution, with additional 3x3 depth-wise convolutions to capture fine details in the high-resolution space. We demonstrate the effectiveness of our LRSA approach by building the LRFormer, a vision transformer with an encoder-decoder structure. Extensive experiments on the ADE20K, COCO-Stuff, and Cityscapes datasets demonstrate that LRFormer outperforms state-of-the-art models. he code is available at https://github.com/yuhuan-wu/LRFormer.
Talking face generation has a wide range of potential applications in the field of virtual digital humans. However, rendering high-fidelity facial video while ensuring lip synchronization is still a challenge for existing audio-driven talking face generation approaches. To address this issue, we propose HyperLips, a two-stage framework consisting of a hypernetwork for controlling lips and a high-resolution decoder for rendering high-fidelity faces. In the first stage, we construct a base face generation network that uses the hypernetwork to control the encoding latent code of the visual face information over audio. First, FaceEncoder is used to obtain latent code by extracting features from the visual face information taken from the video source containing the face frame.Then, HyperConv, which weighting parameters are updated by HyperNet with the audio features as input, will modify the latent code to synchronize the lip movement with the audio. Finally, FaceDecoder will decode the modified and synchronized latent code into visual face content. In the second stage, we obtain higher quality face videos through a high-resolution decoder. To further improve the quality of face generation, we trained a high-resolution decoder, HRDecoder, using face images and detected sketches generated from the first stage as input.Extensive quantitative and qualitative experiments show that our method outperforms state-of-the-art work with more realistic, high-fidelity, and lip synchronization. Project page: https://semchan.github.io/HyperLips Project/