Polyp segmentation is crucial for preventing colorectal cancer a common type of cancer. Deep learning has been used to segment polyps automatically, which reduces the risk of misdiagnosis. Localizing small polyps in colonoscopy images is challenging because of its complex characteristics, such as color, occlusion, and various shapes of polyps. To address this challenge, a novel frequency-based fully convolutional neural network, Multi-Frequency Feature Fusion Polyp Segmentation Network (M3FPolypSegNet) was proposed to decompose the input image into low/high/full-frequency components to use the characteristics of each component. We used three independent multi-frequency encoders to map multiple input images into a high-dimensional feature space. In the Frequency-ASPP Scalable Attention Module (F-ASPP SAM), ASPP was applied between each frequency component to preserve scale information. Subsequently, scalable attention was applied to emphasize polyp regions in a high-dimensional feature space. Finally, we designed three multi-task learning (i.e., region, edge, and distance) in four decoder blocks to learn the structural characteristics of the region. The proposed model outperformed various segmentation models with performance gains of 6.92% and 7.52% on average for all metrics on CVC-ClinicDB and BKAI-IGH-NeoPolyp, respectively.
Vision Transformers (ViTs) have revolutionized medical imaging analysis, showcasing superior efficacy compared to conventional Convolutional Neural Networks (CNNs) in vital tasks such as polyp classification, detection, and segmentation. Leveraging attention mechanisms to focus on specific image regions, ViTs exhibit contextual awareness in processing visual data, culminating in robust and precise predictions, even for intricate medical images. Moreover, the inherent self-attention mechanism in Transformers accommodates varying input sizes and resolutions, granting an unprecedented flexibility absent in traditional CNNs. However, Transformers grapple with challenges like excessive memory usage and limited training parallelism due to self-attention, rendering them impractical for real-time disease detection on resource-constrained devices. In this study, we address these hurdles by investigating the integration of the recently introduced retention mechanism into polyp segmentation, introducing RetSeg, an encoder-decoder network featuring multi-head retention blocks. Drawing inspiration from Retentive Networks (RetNet), RetSeg is designed to bridge the gap between precise polyp segmentation and resource utilization, particularly tailored for colonoscopy images. We train and validate RetSeg for polyp segmentation employing two publicly available datasets: Kvasir-SEG and CVC-ClinicDB. Additionally, we showcase RetSeg's promising performance across diverse public datasets, including CVC-ColonDB, ETIS-LaribPolypDB, CVC-300, and BKAI-IGH NeoPolyp. While our work represents an early-stage exploration, further in-depth studies are imperative to advance these promising findings.
Text is ubiquitous in our visual world, conveying crucial information, such as in documents, websites, and everyday photographs. In this work, we propose UReader, a first exploration of universal OCR-free visually-situated language understanding based on the Multimodal Large Language Model (MLLM). By leveraging the shallow text recognition ability of the MLLM, we only finetuned 1.2% parameters and the training cost is much lower than previous work following domain-specific pretraining and finetuning paradigms. Concretely, UReader is jointly finetuned on a wide range of Visually-situated Language Understanding tasks via a unified instruction format. To enhance the visual text and semantic understanding, we further apply two auxiliary tasks with the same format, namely text reading and key points generation tasks. We design a shape-adaptive cropping module before the encoder-decoder architecture of MLLM to leverage the frozen low-resolution vision encoder for processing high-resolution images. Without downstream finetuning, our single model achieves state-of-the-art ocr-free performance in 8 out of 10 visually-situated language understanding tasks, across 5 domains: documents, tables, charts, natural images, and webpage screenshots. Codes and instruction-tuning datasets will be released.
Generalizable implicit neural representation (INR) enables a single continuous function, i.e., a coordinate-based neural network, to represent multiple data instances by modulating its weights or intermediate features using latent codes. However, the expressive power of the state-of-the-art modulation is limited due to its inability to localize and capture fine-grained details of data entities such as specific pixels and rays. To address this issue, we propose a novel framework for generalizable INR that combines a transformer encoder with a locality-aware INR decoder. The transformer encoder predicts a set of latent tokens from a data instance to encode local information into each latent token. The locality-aware INR decoder extracts a modulation vector by selectively aggregating the latent tokens via cross-attention for a coordinate input and then predicts the output by progressively decoding with coarse-to-fine modulation through multiple frequency bandwidths. The selective token aggregation and the multi-band feature modulation enable us to learn locality-aware representation in spatial and spectral aspects, respectively. Our framework significantly outperforms previous generalizable INRs and validates the usefulness of the locality-aware latents for downstream tasks such as image generation.
Quantum error-correction is a prerequisite for reliable quantum computation. Towards this goal, we present a recurrent, transformer-based neural network which learns to decode the surface code, the leading quantum error-correction code. Our decoder outperforms state-of-the-art algorithmic decoders on real-world data from Google's Sycamore quantum processor for distance 3 and 5 surface codes. On distances up to 11, the decoder maintains its advantage on simulated data with realistic noise including cross-talk, leakage, and analog readout signals, and sustains its accuracy far beyond the 25 cycles it was trained on. Our work illustrates the ability of machine learning to go beyond human-designed algorithms by learning from data directly, highlighting machine learning as a strong contender for decoding in quantum computers.
Recent popularity of Large Language Models (LLMs) has opened countless possibilities in automating numerous AI tasks by connecting LLMs to various domain-specific models or APIs, where LLMs serve as dispatchers while domain-specific models or APIs are action executors. Despite the vast numbers of domain-specific models/APIs, they still struggle to comprehensively cover super diverse automation demands in the interaction between human and User Interfaces (UIs). In this work, we build a multimodal model to ground natural language instructions in given UI screenshots as a generic UI task automation executor. This metadata-free grounding model, consisting of a visual encoder and a language decoder, is first pretrained on well studied document understanding tasks and then learns to decode spatial information from UI screenshots in a promptable way. To facilitate the exploitation of image-to-text pretrained knowledge, we follow the pixel-to-sequence paradigm to predict geometric coordinates in a sequence of tokens using a language decoder. We further propose an innovative Reinforcement Learning (RL) based algorithm to supervise the tokens in such sequence jointly with visually semantic metrics, which effectively strengthens the spatial decoding capability of the pixel-to-sequence paradigm. Extensive experiments demonstrate our proposed reinforced UI instruction grounding model outperforms the state-of-the-art methods by a clear margin and shows the potential as a generic UI task automation API.
We present a universal motion representation that encompasses a comprehensive range of motor skills for physics-based humanoid control. Due to the high dimensionality of humanoids and the inherent difficulties in reinforcement learning, prior methods have focused on learning skill embeddings for a narrow range of movement styles (e.g. locomotion, game characters) from specialized motion datasets. This limited scope hampers their applicability in complex tasks. We close this gap by significantly increasing the coverage of our motion representation space. To achieve this, we first learn a motion imitator that can imitate all of human motion from a large, unstructured motion dataset. We then create our motion representation by distilling skills directly from the imitator. This is achieved by using an encoder-decoder structure with a variational information bottleneck. Additionally, we jointly learn a prior conditioned on proprioception (humanoid's own pose and velocities) to improve model expressiveness and sampling efficiency for downstream tasks. By sampling from the prior, we can generate long, stable, and diverse human motions. Using this latent space for hierarchical RL, we show that our policies solve tasks using human-like behavior. We demonstrate the effectiveness of our motion representation by solving generative tasks (e.g. strike, terrain traversal) and motion tracking using VR controllers.
RGB-T (red–green–blue and thermal) scene parsing has recently drawn considerable research attention. Although existing methods efficiently conduct RGB-T scene parsing, their performance remains limited by a small receptive field. Unlike methods that capture the global context by fusing multiscale features or using an attention mechanism, we propose a graph-enhancement branch network (GEBNet), which uses long-range dependencies obtained from the branch to refine a coarse semantic map generated by the decoder. Semantic and detail modules embedded in the graph-enhancement branch fuse high- and low-level features. Furthermore, inspired by the ability of graph neural networks to capture the global context, we integrate a novel graph-enhancement module into the network branch to obtain global information from both high-level semantic information and low-level details. Results from extensive experiments on the MFNet and PST900 datasets demonstrate the high performance of the proposed GEBNet and the contributions of its main components to the parsing performance.
Lightweight change detection models are essential for industrial applications and edge devices. Reducing the model size while maintaining high accuracy is a key challenge in developing lightweight change detection models. However, many existing methods oversimplify the model architecture, leading to a loss of information and reduced performance. Therefore, developing a lightweight model that can effectively preserve the input information is a challenging problem. To address this challenge, we propose LightCDNet, a novel lightweight change detection model that effectively preserves the input information. LightCDNet consists of an early fusion backbone network and a pyramid decoder for end-to-end change detection. The core component of LightCDNet is the Deep Supervised Fusion Module (DSFM), which guides the early fusion of primary features to improve performance. We evaluated LightCDNet on the LEVIR-CD dataset and found that it achieved comparable or better performance than state-of-the-art models while being 10–117 times smaller in size.
Hand gesture recognition (HGR) has gained significant attention due to the increasing use of AI-powered human-computer interfaces that can interpret the deep spatiotemporal dynamics of biosignals from the peripheral nervous system, such as surface electromyography (sEMG). These interfaces have a range of applications, including the control of extended reality, agile prosthetics, and exoskeletons. However, the natural variability of sEMG among individuals has led researchers to focus on subject-specific solutions. Deep learning methods, which often have complex structures, are particularly data-hungry and can be time-consuming to train, making them less practical for subject-specific applications. In this paper, we propose and develop a generalizable, sequential decoder of transient high-density sEMG (HD-sEMG) that achieves 73% average accuracy on 65 gestures for partially-observed subjects through subject-embedded transfer learning, leveraging pre-knowledge of HGR acquired during pre-training. The use of transient HD-sEMG before gesture stabilization allows us to predict gestures with the ultimate goal of counterbalancing system control delays. The results show that the proposed generalized models significantly outperform subject-specific approaches, especially when the training data is limited, and there is a significant number of gesture classes. By building on pre-knowledge and incorporating a multiplicative subject-embedded structure, our method comparatively achieves more than 13% average accuracy across partially observed subjects with minimal data availability. This work highlights the potential of HD-sEMG and demonstrates the benefits of modeling common patterns across users to reduce the need for large amounts of data for new users, enhancing practicality.
We consider low-latency image transmission over a noisy wireless channel when correlated side information is present only at the receiver side (the Wyner-Ziv scenario). In particular, we are interested in developing practical schemes using a data-driven joint source-channel coding (JSCC) approach, which has been previously shown to outperform conventional separation-based approaches in the practical finite blocklength regimes, and to provide graceful degradation with channel quality. We propose a novel neural network architecture that incorporates the decoder-only side information at multiple stages at the receiver side. Our results demonstrate that the proposed method succeeds in integrating the side information, yielding improved performance at all channel conditions in terms of the various quality measures considered here, especially at low channel signal-to-noise ratios (SNRs) and small bandwidth ratios (BRs). We have made the source code of the proposed method public to enable further research, and the reproducibility of the results.
Masked Autoencoders (MAE) have demonstrated promising performance in self-supervised learning for both 2D and 3D computer vision. Nevertheless, existing MAE-based methods still have certain drawbacks. Firstly, the functional decoupling between the encoder and decoder is incomplete, which limits the encoder's representation learning ability. Secondly, downstream tasks solely utilize the encoder, failing to fully leverage the knowledge acquired through the encoder-decoder architecture in the pre-text task. In this paper, we propose Point Regress AutoEncoder (Point-RAE), a new scheme for regressive autoencoders for point cloud self-supervised learning. The proposed method decouples functions between the decoder and the encoder by introducing a mask regressor, which predicts the masked patch representation from the visible patch representation encoded by the encoder and the decoder reconstructs the target from the predicted masked patch representation. By doing so, we minimize the impact of decoder updates on the representation space of the encoder. Moreover, we introduce an alignment constraint to ensure that the representations for masked patches, predicted from the encoded representations of visible patches, are aligned with the masked patch presentations computed from the encoder. To make full use of the knowledge learned in the pre-training stage, we design a new finetune mode for the proposed Point-RAE. Extensive experiments demonstrate that our approach is efficient during pre-training and generalizes well on various downstream tasks. Specifically, our pre-trained models achieve a high accuracy of \textbf{90.28\%} on the ScanObjectNN hardest split and \textbf{94.1\%} accuracy on ModelNet40, surpassing all the other self-supervised learning methods. Our code and pretrained model are public available at: \url{https://github.com/liuyyy111/Point-RAE}.
This paper introduces the T23 team's system submitted to the Singing Voice Conversion Challenge 2023. Following the recognition-synthesis framework, our singing conversion model is based on VITS, incorporating four key modules: a prior encoder, a posterior encoder, a decoder, and a parallel bank of transposed convolutions (PBTC) module. We particularly leverage Whisper, a powerful pre-trained ASR model, to extract bottleneck features (BNF) as the input of the prior encoder. Before BNF extraction, we perform pitch perturbation to the source signal to remove speaker timbre, which effectively avoids the leakage of the source speaker timbre to the target. Moreover, the PBTC module extracts multi-scale F0 as the auxiliary input to the prior encoder, thereby capturing better pitch variations of singing. We design a three-stage training strategy to better adapt the base model to the target speaker with limited target speaker data. Official challenge results show that our system has superior performance in naturalness, ranking 1st and 2nd respectively in Task 1 and 2. Further ablation justifies the effectiveness of our system design.
Generalized quadrature spatial modulation (GQSM) schemes are known to achieve high energy- and spectral- efficiencies by modulating information both in transmitted symbols and in coded combinatorial activations of subsets of multiple transmit antennas. A challenge of the approach is, however, the decoding complexity which scales with the efficiency of the scheme. In order to circumvent this bottleneck and enable high-performance and feasible GQSM in massive multiple-input multiple-output (mMIMO) scenarios, we propose a novel decoding algorithm which enjoys a complexity order that is independent of the combinatorial factor. This remarkable feature of the proposed decoder is a consequence of a novel vectorized Gaussian belief propagation (GaBP) algorithm, here contributed, whose message passing (MP) rules leverage both pilot symbols and the unit vector decomposition (UVD) of the GQSM signal structure. The effectiveness of the proposed UVD-GaBP method is illustrated via computer simulations including numerical results for systems of a size never before reported in related literature (up to 32 transmit antennas), which demonstrates the potential of the approach in paving the way towards high energy and spectral efficiency for wireless systems in a truly mMIMO setting.
In recent years, many interpretability methods have been proposed to help interpret the internal states of Transformer-models, at different levels of precision and complexity. Here, to analyze encoder-decoder Transformers, we propose a simple, new method: DecoderLens. Inspired by the LogitLens (for decoder-only Transformers), this method involves allowing the decoder to cross-attend representations of intermediate encoder layers instead of using the final encoder output, as is normally done in encoder-decoder models. The method thus maps previously uninterpretable vector representations to human-interpretable sequences of words or symbols. We report results from the DecoderLens applied to models trained on question answering, logical reasoning, speech recognition and machine translation. The DecoderLens reveals several specific subtasks that are solved at low or intermediate layers, shedding new light on the information flow inside the encoder component of this important class of models.
Variational dimensionality reduction methods are widely used for their accuracy, generative capabilities, and robustness. We introduce a unifying framework that generalizes both such as traditional and state-of-the-art methods. The framework is based on an interpretation of the multivariate information bottleneck, trading off the information preserved in an encoder graph (defining what to compress) against that in a decoder graph (defining a generative model for data). Using this approach, we rederive existing methods, including the deep variational information bottleneck, variational autoencoders, and deep multiview information bottleneck. We naturally extend the deep variational CCA (DVCCA) family to beta-DVCCA and introduce a new method, the deep variational symmetric information bottleneck (DVSIB). DSIB, the deterministic limit of DVSIB, connects to modern contrastive learning approaches such as Barlow Twins, among others. We evaluate these methods on Noisy MNIST and Noisy CIFAR-100, showing that algorithms better matched to the structure of the problem like DVSIB and beta-DVCCA produce better latent spaces as measured by classification accuracy, dimensionality of the latent variables, sample efficiency, and consistently outperform other approaches under comparable conditions. Additionally, we benchmark against state-of-the-art models, achieving superior or competitive accuracy. Our results demonstrate that this framework can seamlessly incorporate diverse multi-view representation learning algorithms, providing a foundation for designing novel, problem-specific loss functions.
Ultrasound imaging serves as an effective and non-invasive diagnostic tool commonly employed in clinical examinations. However, the presence of speckle noise in ultrasound images invariably degrades image quality, impeding the performance of subsequent tasks, such as segmentation and classification. Existing methods for speckle noise reduction frequently induce excessive image smoothing or fail to preserve detailed information adequately. In this paper, we propose a complementary global and local knowledge network for ultrasound denoising with fine-grained refinement. Initially, the proposed architecture employs the L-CSwinTransformer as encoder to capture global information, incorporating CNN as decoder to fuse local features. We expand the resolution of the feature at different stages to extract more global information compared to the original CSwinTransformer. Subsequently, we integrate Fine-grained Refinement Block (FRB) within the skip-connection stage to further augment features. We validate our model on two public datasets, HC18 and BUSI. Experimental results demonstrate that our model can achieve competitive performance in both quantitative metrics and visual performance. Our code will be available at https://github.com/AAlkaid/USDenoising.
Many recent breakthroughs in machine learning have been enabled by the pre-trained foundation models. By scaling up model parameters, training data, and computation resources, foundation models have significantly advanced the state-of-the-art in many applications. However, it is still an open question of how to use these models to perform downstream tasks efficiently. Knowledge distillation (KD) has been explored to tackle this challenge. KD transfers knowledge from a large teacher model to a smaller student model. While KD has been successful in improving student model performance, recent research has discovered that a powerful teacher does not necessarily lead to a powerful student, due to their huge capacity gap. In addition, the potential distribution shifts between the pre-training data and downstream tasks can make knowledge transfer in KD sub-optimal for improving downstream task performance. In this paper, we extend KD with an interactive communication process to help students of downstream tasks learn effectively from pre-trained foundation models. Our design is inspired by the way humans learn from teachers who can explain knowledge in a way that meets the students' needs. Specifically, we let each model (i.e., student and teacher) train two components: (1) an encoder encoding the model's hidden states to a message and (2) a decoder decoding any messages to its own hidden states. With encoder and decoder, not only can the teacher transfer rich information by encoding its hidden states, but also the student can send messages with information of downstream tasks to the teacher. Therefore, knowledge passing from teacher to student can be tailored to the student's capacity and downstream tasks' distributions. We conducted experiments on benchmark datasets to show that our communication mechanism outperforms state-of-the-art distillation techniques.
Prompting and adapter tuning have emerged as efficient alternatives to fine-tuning (FT) methods. However, existing studies on speech prompting focused on classification tasks and failed on more complex sequence generation tasks. Besides, adapter tuning is primarily applied with a focus on encoder-only self-supervised models. Our experiments show that prompting on Wav2Seq, a self-supervised encoder-decoder model, surpasses previous works in sequence generation tasks. It achieves a remarkable 53% relative improvement in word error rate for ASR and a 27% in F1 score for slot filling. Additionally, prompting competes with the FT method in the low-resource scenario. Moreover, we show the transferability of prompting and adapter tuning on Wav2Seq in cross-lingual ASR. When limited trainable parameters are involved, prompting and adapter tuning consistently outperform conventional FT across 7 languages. Notably, in the low-resource scenario, prompting consistently outperforms adapter tuning.
Designing novel proteins with desired functions is crucial in biology and chemistry. However, most existing work focus on protein sequence design, leaving protein sequence and structure co-design underexplored. In this paper, we propose GeoPro, a method to design protein backbone structure and sequence jointly. Our motivation is that protein sequence and its backbone structure constrain each other, and thus joint design of both can not only avoid nonfolding and misfolding but also produce more diverse candidates with desired functions. To this end, GeoPro is powered by an equivariant encoder for three-dimensional (3D) backbone structure and a protein sequence decoder guided by 3D geometry. Experimental results on two biologically significant metalloprotein datasets, including $\beta$-lactamases and myoglobins, show that our proposed GeoPro outperforms several strong baselines on most metrics. Remarkably, our method discovers novel $\beta$-lactamases and myoglobins which are not present in protein data bank (PDB) and UniProt. These proteins exhibit stable folding and active site environments reminiscent of those of natural proteins, demonstrating their excellent potential to be biologically functional.