Deep learning has made a breakthrough in medical image segmentation in recent years due to its ability to extract high-level features without the need for prior knowledge. In this context, U-Net is one of the most advanced medical image segmentation models, with promising results in mammography. Despite its excellent overall performance in segmenting multimodal medical images, the traditional U-Net structure appears to be inadequate in various ways. There are certain U-Net design modifications, such as MultiResUNet, Connected-UNets, and AU-Net, that have improved overall performance in areas where the conventional U-Net architecture appears to be deficient. Following the success of UNet and its variants, we have presented two enhanced versions of the Connected-UNets architecture: ConnectedUNets+ and ConnectedUNets++. In ConnectedUNets+, we have replaced the simple skip connections of Connected-UNets architecture with residual skip connections, while in ConnectedUNets++, we have modified the encoder-decoder structure along with employing residual skip connections. We have evaluated our proposed architectures on two publicly available datasets, the Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM) and INbreast.
Colorectal cancer (CRC) is the second leading cause of cancer-related death worldwide. Excision of polyps during colonoscopy helps reduce mortality and morbidity for CRC. Powered by deep learning, computer-aided diagnosis (CAD) systems can detect regions in the colon overlooked by physicians during colonoscopy. Lacking high accuracy and real-time speed are the essential obstacles to be overcome for successful clinical integration of such systems. While literature is focused on improving accuracy, the speed parameter is often ignored. Toward this critical need, we intend to develop a novel real-time deep learning-based architecture, DilatedSegNet, to perform polyp segmentation on the fly. DilatedSegNet is an encoder-decoder network that uses pre-trained ResNet50 as the encoder from which we extract four levels of feature maps. Each of these feature maps is passed through a dilated convolution pooling (DCP) block. The outputs from the DCP blocks are concatenated and passed through a series of four decoder blocks that predicts the segmentation mask. The proposed method achieves a real-time operation speed of 33.68 frames per second with an average dice coefficient of 0.90 and mIoU of 0.83. Additionally, we also provide heatmap along with the qualitative results that shows the explanation for the polyp location, which increases the trustworthiness of the method. The results on the publicly available Kvasir-SEG and BKAI-IGH datasets suggest that DilatedSegNet can give real-time feedback while retaining a high \ac{DSC}, indicating high potential for using such models in real clinical settings in the near future. The GitHub link of the source code can be found here: \url{https://github.com/nikhilroxtomar/DilatedSegNet}.
Rationalization is fundamental to human reasoning and learning. NLP models trained to produce rationales along with predictions, called self-rationalization models, have been investigated for their interpretability and utility to end-users. However, the extent to which training with human-written rationales facilitates learning remains an under-explored question. We ask whether training models to self-rationalize can aid in their learning to solve tasks for the right reasons. Specifically, we evaluate how training self-rationalization models with free-text rationales affects robustness to spurious correlations in fine-tuned encoder-decoder and decoder-only models of six different sizes. We evaluate robustness to spurious correlations by measuring performance on 1) manually annotated challenge datasets and 2) subsets of original test sets where reliance on spurious correlations would fail to produce correct answers. We find that while self-rationalization can improve robustness to spurious correlations in low-resource settings, it tends to hurt robustness in higher-resource settings. Furthermore, these effects depend on model family and size, as well as on rationale content. Together, our results suggest that explainability can come at the cost of robustness; thus, appropriate care should be taken when training self-rationalizing models with the goal of creating more trustworthy models.
Depth information is the foundation of perception, essential for autonomous driving, robotics, and other source-constrained applications. Promptly obtaining accurate and efficient depth information allows for a rapid response in dynamic environments. Sensor-based methods using LIDAR and RADAR obtain high precision at the cost of high power consumption, price, and volume. While due to advances in deep learning, vision-based approaches have recently received much attention and can overcome these drawbacks. In this work, we explore an extreme scenario in vision-based settings: estimate a depth map from one monocular image severely plagued by grid artifacts and blurry edges. To address this scenario, We first design a convolutional attention mechanism block (CAMB) which consists of channel attention and spatial attention sequentially and insert these CAMBs into skip connections. As a result, our novel approach can find the focus of current image with minimal overhead and avoid losses of depth features. Next, by combining the depth value, the gradients of X axis, Y axis and diagonal directions, and the structural similarity index measure (SSIM), we propose our novel loss function. Moreover, we utilize pixel blocks to accelerate the computation of the loss function. Finally, we show, through comprehensive experiments on two large-scale image datasets, i.e. KITTI and NYU-V2, that our method outperforms several representative baselines.
Conventional autoregressive left-to-right (L2R) sequence generation faces two issues during decoding: limited to unidirectional target sequence modeling, and constrained on strong local dependencies. To address the aforementioned problem, we propose P$^3$LM, a probabilistically permuted prophet language model, which strengthens the modeling of bidirectional information and long token dependencies for sequence generation. Specifically, P$^3$LM learns to generate tokens in permuted order upon an order-aware transformer decoder, as well as to generate the corresponding future $N$ tokens with a multi-stream attention mechanism. Extensive experiments are conducted on the GLGE benchmark, which includes four datasets for summarization, two for question generation, one for conversational question answering, and one for dialog response generation, where P$^3$LM achieves state-of-the-art results compared with strong publicly available generative pre-training methods.
A straightforward approach to context-aware neural machine translation consists in feeding the standard encoder-decoder architecture with a window of consecutive sentences, formed by the current sentence and a number of sentences from its context concatenated to it. In this work, we propose an improved concatenation approach that encourages the model to focus on the translation of the current sentence, discounting the loss generated by target context. We also propose an additional improvement that strengthen the notion of sentence boundaries and of relative sentence distance, facilitating model compliance to the context-discounted objective. We evaluate our approach with both average-translation quality metrics and contrastive test sets for the translation of inter-sentential discourse phenomena, proving its superiority to the vanilla concatenation approach and other sophisticated context-aware systems.
This study presents a novel unsupervised convolutional Neural Network (NN) architecture with nonlocal interactions for solving Partial Differential Equations (PDEs). The nonlocal Peridynamic Differential Operator (PDDO) is employed as a convolutional filter for evaluating derivatives the field variable. The NN captures the time-dynamics in smaller latent space through encoder-decoder layers with a Convolutional Long-short Term Memory (ConvLSTM) layer between them. The ConvLSTM architecture is modified by employing a novel activation function to improve the predictive capability of the learning architecture for physics with periodic behavior. The physics is invoked in the form of governing equations at the output of the NN and in the latent (reduced) space. By considering a few benchmark PDEs, we demonstrate the training performance and extrapolation capability of this novel NN architecture by comparing against Physics Informed Neural Networks (PINN) type solvers. It is more capable of extrapolating the solution for future timesteps than the other existing architectures.
Cancer of the brain is deadly and requires careful surgical segmentation. The brain tumors were segmented using U-Net using a Convolutional Neural Network (CNN). When looking for overlaps of necrotic, edematous, growing, and healthy tissue, it might be hard to get relevant information from the images. The 2D U-Net network was improved and trained with the BraTS datasets to find these four areas. U-Net can set up many encoder and decoder routes that can be used to get information from images that can be used in different ways. To reduce computational time, we use image segmentation to exclude insignificant background details. Experiments on the BraTS datasets show that our proposed model for segmenting brain tumors from MRI (MRI) works well. In this study, we demonstrate that the BraTS datasets for 2017, 2018, 2019, and 2020 do not significantly differ from the BraTS 2019 dataset's attained dice scores of 0.8717 (necrotic), 0.9506 (edema), and 0.9427 (enhancing).
The usage of automatic speech recognition (ASR) systems are becoming omnipresent ranging from personal assistant to chatbots, home, and industrial automation systems, etc. Modern robots are also equipped with ASR capabilities for interacting with humans as speech is the most natural interaction modality. However, ASR in robots faces additional challenges as compared to a personal assistant. Being an embodied agent, a robot must recognize the physical entities around it and therefore reliably recognize the speech containing the description of such entities. However, current ASR systems are often unable to do so due to limitations in ASR training, such as generic datasets and open-vocabulary modeling. Also, adverse conditions during inference, such as noise, accented, and far-field speech makes the transcription inaccurate. In this work, we present a method to incorporate a robot's visual information into an ASR system and improve the recognition of a spoken utterance containing a visible entity. Specifically, we propose a new decoder biasing technique to incorporate the visual context while ensuring the ASR output does not degrade for incorrect context. We achieve a 59% relative reduction in WER from an unmodified ASR system.
We propose to directly compute classification estimates by learning features encoded with their class scores using PCA. Our resulting model has a encoder-decoder structure suitable for supervised learning, it is computationally efficient and performs well for classification on several datasets.
We introduce a state-of-the-art real-time, high-fidelity, audio codec leveraging neural networks. It consists in a streaming encoder-decoder architecture with quantized latent space trained in an end-to-end fashion. We simplify and speed-up the training by using a single multiscale spectrogram adversary that efficiently reduces artifacts and produce high-quality samples. We introduce a novel loss balancer mechanism to stabilize training: the weight of a loss now defines the fraction of the overall gradient it should represent, thus decoupling the choice of this hyper-parameter from the typical scale of the loss. Finally, we study how lightweight Transformer models can be used to further compress the obtained representation by up to 40%, while staying faster than real time. We provide a detailed description of the key design choices of the proposed model including: training objective, architectural changes and a study of various perceptual loss functions. We present an extensive subjective evaluation (MUSHRA tests) together with an ablation study for a range of bandwidths and audio domains, including speech, noisy-reverberant speech, and music. Our approach is superior to the baselines methods across all evaluated settings, considering both 24 kHz monophonic and 48 kHz stereophonic audio. Code and models are available at github.com/facebookresearch/encodec.
The segmentation of histopathological whole slide images into tumourous and non-tumourous types of tissue is a challenging task that requires the consideration of both local and global spatial contexts to classify tumourous regions precisely. The identification of subtypes of tumour tissue complicates the issue as the sharpness of separation decreases and the pathologist's reasoning is even more guided by spatial context. However, the identification of detailed types of tissue is crucial for providing personalized cancer therapies. Due to the high resolution of whole slide images, existing semantic segmentation methods, restricted to isolated image sections, are incapable of processing context information beyond. To take a step towards better context comprehension, we propose a patch neighbour attention mechanism to query the neighbouring tissue context from a patch embedding memory bank and infuse context embeddings into bottleneck hidden feature maps. Our memory attention framework (MAF) mimics a pathologist's annotation procedure -- zooming out and considering surrounding tissue context. The framework can be integrated into any encoder-decoder segmentation method. We evaluate the MAF on a public breast cancer and an internal kidney cancer data set using famous segmentation models (U-Net, DeeplabV3) and demonstrate the superiority over other context-integrating algorithms -- achieving a substantial improvement of up to $17\%$ on Dice score. The code is publicly available at: https://github.com/tio-ikim/valuing-vicinity
We present a unified and compact scene representation for robotics, where each object in the scene is depicted by a latent code capturing geometry and appearance. This representation can be decoded for various tasks such as novel view rendering, 3D reconstruction (e.g. recovering depth, point clouds, or voxel maps), collision checking, and stable grasp prediction. We build our representation from a single RGB input image at test time by leveraging recent advances in Neural Radiance Fields (NeRF) that learn category-level priors on large multiview datasets, then fine-tune on novel objects from one or few views. We expand the NeRF model for additional grasp outputs and explore ways to leverage this representation for robotics. At test-time, we build the representation from a single RGB input image observing the scene from only one viewpoint. We find that the recovered representation allows rendering from novel views, including of occluded object parts, and also for predicting successful stable grasps. Grasp poses can be directly decoded from our latent representation with an implicit grasp decoder. We experimented in both simulation and real world and demonstrated the capability for robust robotic grasping using such compact representation. Website: https://nerfgrasp.github.io
Neural retrieval models have superseded classic bag-of-words methods such as BM25 as the retrieval framework of choice. However, neural systems lack the interpretability of bag-of-words models; it is not trivial to connect a query change to a change in the latent space that ultimately determines the retrieval results. To shed light on this embedding space, we learn a "query decoder" that, given a latent representation of a neural search engine, generates the corresponding query. We show that it is possible to decode a meaningful query from its latent representation and, when moving in the right direction in latent space, to decode a query that retrieves the relevant paragraph. In particular, the query decoder can be useful to understand "what should have been asked" to retrieve a particular paragraph from the collection. We employ the query decoder to generate a large synthetic dataset of query reformulations for MSMarco, leading to improved retrieval performance. On this data, we train a pseudo-relevance feedback (PRF) T5 model for the application of query suggestion that outperforms both query reformulation and PRF information retrieval baselines.
Encoder-decoder architecture is widely adopted for sequence-to-sequence modeling tasks. For machine translation, despite the evolution from long short-term memory networks to Transformer networks, plus the introduction and development of attention mechanism, encoder-decoder is still the de facto neural network architecture for state-of-the-art models. While the motivation for decoding information from some hidden space is straightforward, the strict separation of the encoding and decoding steps into an encoder and a decoder in the model architecture is not necessarily a must. Compared to the task of autoregressive language modeling in the target language, machine translation simply has an additional source sentence as context. Given the fact that neural language models nowadays can already handle rather long contexts in the target language, it is natural to ask whether simply concatenating the source and target sentences and training a language model to do translation would work. In this work, we investigate the aforementioned concept for machine translation. Specifically, we experiment with bilingual translation, translation with additional target monolingual data, and multilingual translation. In all cases, this alternative approach performs on par with the baseline encoder-decoder Transformer, suggesting that an encoder-decoder architecture might be redundant for neural machine translation.
In recent years, hyperspectral image classification (HSIC) has achieved impressive progress with emerging studies on deep learning models. However, the classification performance downgrades due to the limited number of annotated samples, especially for minority classes. Notably, the imbalanced data dilemma is familiar in remote sensing hyperspectral image because the ground objects are commonly distributed without evenness. Therefore, this paper proposes a novel deep generative spectral-spatial classifier (DGSSC) for addressing the issues of imbalanced HSIC. Specifically, the DGSSC comprises three components, a two-stage encoder, a decoder, and a classifier, which are trained in an end-to-end manner. In particular, to exploit the abundant spectral-spatial features with relatively low computational complexity, the first stage of the encoder comprises successive three-dimensional (3D) and two-dimensional (2D) convolutions, exploring the spectral-spatial and deep spatial information. In addition, the second stage involves the deep latent variable model to achieve minority-class data augmentation. Furthermore, a patch distance-based reconstruction loss function is designed to facilitate the outputs of the decoder being more similar to the input 3D patch samples. The proposed DGSSC can outperform the state-of-the-art methods on three benchmark datasets, especially with its more robust prediction results. For instance, the DGSSC achieves a remarkable 97.85% mean overall accuracy with 0.24% standard deviation over ten independent runs with randomly selected imbalanced 1% training samples on the University of Pavia dataset.
Model-based and learning-based methods are two major types of methodologies to model car following behaviors. Model-based methods describe the car-following behaviors with explicit mathematical equations, while learning-based methods focus on getting a mapping between inputs and outputs. Both types of methods have advantages and weaknesses. Meanwhile, most car-following models are generative and only consider the inputs of the speed, position, and acceleration of the last time step. To address these issues, this study proposes a novel framework called IDM-Follower that can generate a sequence of following vehicle trajectory by a recurrent autoencoder informed by a physical car-following model, the Intelligent Driving Model (IDM).We implement a novel structure with two independent encoders and a self-attention decoder that could sequentially predict the following trajectories. A loss function considering the discrepancies between predictions and labeled data integrated with discrepancies from model-based predictions is implemented to update the neural network parameters. Numerical experiments with multiple settings on simulation and NGSIM datasets show that the IDM-Follower can improve the prediction performance compared to the model-based or learning-based methods alone. Analysis on different noise levels also shows good robustness of the model.
Synthetic image generation has recently experienced significant improvements in domains such as natural image or art generation. However, the problem of figure and diagram generation remains unexplored. A challenging aspect of generating figures and diagrams is effectively rendering readable texts within the images. To alleviate this problem, we present OCR-VQGAN, an image encoder, and decoder that leverages OCR pre-trained features to optimize a text perceptual loss, encouraging the architecture to preserve high-fidelity text and diagram structure. To explore our approach, we introduce the Paper2Fig100k dataset, with over 100k images of figures and texts from research papers. The figures show architecture diagrams and methodologies of articles available at arXiv.org from fields like artificial intelligence and computer vision. Figures usually include text and discrete objects, e.g., boxes in a diagram, with lines and arrows that connect them. We demonstrate the effectiveness of OCR-VQGAN by conducting several experiments on the task of figure reconstruction. Additionally, we explore the qualitative and quantitative impact of weighting different perceptual metrics in the overall loss function. We release code, models, and dataset at https://github.com/joanrod/ocr-vqgan.
Image-to-text tasks, such as open-ended image captioning and controllable image description, have received extensive attention for decades. Here, we further advance this line of work by presenting Visual Spatial Description (VSD), a new perspective for image-to-text toward spatial semantics. Given an image and two objects inside it, VSD aims to produce one description focusing on the spatial perspective between the two objects. Accordingly, we manually annotate a dataset to facilitate the investigation of the newly-introduced task and build several benchmark encoder-decoder models by using VL-BART and VL-T5 as backbones. In addition, we investigate pipeline and joint end-to-end architectures for incorporating visual spatial relationship classification (VSRC) information into our model. Finally, we conduct experiments on our benchmark dataset to evaluate all our models. Results show that our models are impressive, providing accurate and human-like spatial-oriented text descriptions. Meanwhile, VSRC has great potential for VSD, and the joint end-to-end architecture is the better choice for their integration. We make the dataset and codes public for research purposes.
Overlapping community detection is a key problem in graph mining. Some research has considered applying graph convolutional networks (GCN) to tackle the problem. However, it is still challenging to incorporate deep graph convolutional networks in the case of general irregular graphs. In this study, we design a deep dynamic residual graph convolutional network (DynaResGCN) based on our novel dynamic dilated aggregation mechanisms and a unified end-to-end encoder-decoder-based framework to detect overlapping communities in networks. The deep DynaResGCN model is used as the encoder, whereas we incorporate the Bernoulli-Poisson (BP) model as the decoder. Consequently, we apply our overlapping community detection framework in a research topics dataset without having ground truth, a set of networks from Facebook having a reliable (hand-labeled) ground truth, and in a set of very large co-authorship networks having empirical (not hand-labeled) ground truth. Our experimentation on these datasets shows significantly superior performance over many state-of-the-art methods for the detection of overlapping communities in networks.