Single-cell RNA-seq data has become a critical source in revealing cellular activities. However, the developing probing techniques and the fatal damages to the detected cells incur various kinds of noise, e.g. the batch effect and the absence of cellular correspondence between experimental groups. Therefore, many single-cell tasks are better modeled as generative rather than discriminative tasks, since, instead of the exact cell-wise ground truth, only the distribution of cellular profiles under a certain condition is measurable. Considering the highly nonlinear and complex associations between gene expressions, we developed scVDM, a latent diffusion model integrated with a transformer-based conditional denoiser to learn three different generative tasks in single-cell data, including conditional data generation, batch effect correction and drug perturbation prediction. The high dimensional transcriptomic data are firstly projected to the latent space through a conditional VAE and then the complicated relationships between latent dimensions are deeply exploited through self-attentions to generate realistic diffusion noise. Based on the evaluation of five real-world datasets, our method demonstrates outstanding performance through comprehensive experimental results in all generative tasks.

The poor generalizability of survival prediction models, stemming from the high-dimensional and heterogeneous nature of transcriptomic data, severely limits their clinical application. A promising strategy is to incorporate prior knowledge, yet this presents a dilemma: knowledge from biological databases (e.g., Reactome) often mismatches the complexity of clinical tumors, while data-driven knowledge (e.g., gene correlations) is confounded by cohort-specific noise. To address this challenge, we propose SynerGNN, a Synergistic Graph Neural Network that intelligently fuses these two imperfect yet complementary knowledge sources. SynerGNN features a dual-stream graph construction process. One stream embeds robust topological structures from the Reactome knowledgebase. In parallel, the other stream adaptively builds a data-driven graph from cohortspecific expression patterns after prognostic biomarker selection. Subsequently, a dynamic attention mechanism fuses the representations from these two streams, learning to dynamically balance the contributions of the stable biological prior and the context-specific data evidence. Evaluations on multiple public cancer datasets show that SynerGNN achieves a significant average improvement in the concordance index (C -index) and demonstrates superior robustness in cross-cohort validation. Our work provides a novel framework that effectively navigates the limitations of different prior knowledge sources, offering a generalizable solution for transcriptomic survival analysis.

Studying the outcomes of genetic perturbation based on single-cell RNA-seq data is crucial for understanding genetic regulation of cells. However, the high cost of cellular experiments and single-cell sequencing restrict us from measuring the full combination space of genetic perturbations and cell types. Consequently, a bunch of computational models have been proposed to predict unseen combinations based on existing data. Among them, generative models, e.g. variational autoencoder and diffusion models, have the superiority in capturing the perturbed data distribution, but lack a biologically understandable foundation for generalization. On the other side of the spectrum, Gene Regulation Networks or gene pathway knowledge have been exploited for more reasonable generalization enhancement. Unfortunately, they do not reach a balanced processing of the two data modalities, leading to a degraded fitting ability. Hence, we propose a dual-stream architecture. Before the information from two modalities are merged, the sequencing data are learned with a generative model while three types of knowledge data are comprehensively processed with graph networks and a masked transformer, enforcing a deep understanding of single-modality data, respectively. The benchmark results show an approximate 20% reduction in terms of mean squared error, proving the effectiveness of the model.

Accurate and efficient classification of electrocardiogram (ECG) signals is an important task in clinical diagnosis and patient care. Existing approaches of ECG signal classification suffer from problems such as feature redundancy, overfitting, increased computational complexity, and decreased real-time performance. The objective of this study is to propose an effective and efficient solution for ECG signal classification. We introduce a novel end-to-end multi-label classification model named MSR-D2CNN that combines a multi-scale dynamic depthwise convolution module (MSD2C-Blocks) with residual skip connections. MSD2C-Blocks integrate the channel attention mechanism Squeeze-and-Excitation(SE) with the ECG dynamic convolution (EDC) module. The model can capture features of different sizes and shapes of ECG signals while maintaining a lightweight structure. We evaluate the proposed model on the PTB-XL dataset, which consists of 12-lead ECG signals and covers 6 different label settings. Our proposed model achieves excellent classification performance while maintaining a low parameter count and computational complexity. Specifically, the parameter count is 0.827M, FLOPs is 134.563M, and Macro-AUC scores are 0.9350, 0.9360, 0.9339, 0.9341, 0.8765, and 0.9668 across the six different label settings. Our proposed solution provides an effective and efficient approach to ECG signal classification, overcoming the limitations of existing methods. The MSR-D2CNN model can accurately classify ECG signals while maintaining a low parameter count and computational complexity, making it practical for real-world scenarios.

De novo peptide sequencing is a promising approach for novel peptide discovery, highlighting the performance improvements for the state-of-the-art models. The quality of mass spectra often varies due to unexpected missing of certain ions, presenting a significant challenge in de novo peptide sequencing. Here, we use a novel concept of complementary spectra to enhance ion information of the experimental spectrum and demonstrate it through conceptual and practical analyses. Afterward, we design suitable encoders to encode the experimental spectrum and the corresponding complementary spectrum and propose a de novo sequencing model $\pi$-HelixNovo based on the Transformer architecture. We first demonstrated that $\pi$-HelixNovo outperforms other state-of-the-art models using a series of comparative experiments. Then, we utilized $\pi$-HelixNovo to de novo gut metaproteome peptides for the first time. The results show $\pi$-HelixNovo increases the identification coverage and accuracy of gut metaproteome and enhances the taxonomic resolution of gut metaproteome. We finally trained a powerful $\pi$-HelixNovo utilizing a larger training dataset, and as expected, $\pi$-HelixNovo achieves unprecedented performance, even for peptide-spectrum matches with never-before-seen peptide sequences. We also use the powerful $\pi$-HelixNovo to identify antibody peptides and multi-enzyme cleavage peptides, and $\pi$-HelixNovo is highly robust in these applications. Our results demonstrate the effectivity of the complementary spectrum and take a significant step forward in de novo peptide sequencing.

Transcriptome-based survival modeling is a critical yet complicated task in cancer treatment due to the strong association between the prognosis of a patient and the progression of tumors based on heterogeneous molecular mechanisms. An effective prognosis model needs to consider the uncertainty that naturally stems from the high molecular heterogeneity together with numerous unforeseeable factors related to prognosis, and requires a perfect match between the selected covariates and the modeling approach of the joint distribution between covariates and right-censored survival time. Unfortunately, these necessary factors have not been fully considered in existing survival models. In this paper, we propose a novel conditional variational deep survival model (CVaDeS for short) for survival analysis. CVaDeS accurately predicts the risk values of patients and effectively distinguishes high-risk and low-risk individuals. Also, the proposed model CVaDeS can be effectively combined with Cox feature selection for survival analysis. Extensive experimental results demonstrate that the combination of Cox feature selection with CVaDeS significantly improves the predictive performance of the model, surpassing baseline models.

We study the problem of efficient semantic segmentation of large-scale 3D point clouds. By relying on expensive sampling techniques or computationally heavy pre/post-processing steps, most existing approaches are only able to be trained and operate over small-scale point clouds. In this paper, we introduce RandLA-Net, an efficient and lightweight neural architecture to directly infer per-point semantics for large-scale point clouds. The key to our approach is to use random point sampling instead of more complex point selection approaches. Although remarkably computation and memory efficient, random sampling can discard key features by chance. To overcome this, we introduce a novel local feature aggregation module to progressively increase the receptive field for each 3D point, thereby effectively preserving geometric details. Comparative experiments show that our RandLA-Net can process 1 million points in a single pass up to 200× faster than existing approaches. Moreover, extensive experiments on five large-scale point cloud datasets, including Semantic3D, SemanticKITTI, Toronto3D, NPM3D and S3DIS, demonstrate the state-of-the-art semantic segmentation performance of our RandLA-Net.

Due to the sparse rewards and high degree of environmental variation, reinforcement learning approaches, such as deep deterministic policy gradient (DDPG), are plagued by issues of high variance when applied in complex real-world environments. We present a new framework for overcoming these issues by incorporating a stochastic switch, allowing an agent to choose between high- and low-variance policies. The stochastic switch can be jointly trained with the original DDPG in the same framework. In this article, we demonstrate the power of the framework in a navigation task, where the robot can dynamically choose to learn through exploration or to use the output of a heuristic controller as guidance. Instead of starting from completely random actions, the navigation capability of a robot can be quickly bootstrapped by several simple independent controllers. The experimental results show that with the aid of stochastic guidance, we are able to effectively and efficiently train DDPG navigation policies and achieve significantly better performance than state-of-the-art baseline models.

We study the problem of efficient semantic segmentation for large-scale 3D point clouds. By relying on expensive sampling techniques or computationally heavy pre/post-processing steps, most existing approaches are only able to be trained and operate over small-scale point clouds. In this paper, we introduce RandLA-Net, an efficient and lightweight neural architecture to directly infer per-point semantics for large-scale point clouds. The key to our approach is to use random point sampling instead of more complex point selection approaches. Although remarkably computation and memory efficient, random sampling can discard key features by chance. To overcome this, we introduce a novel local feature aggregation module to progressively increase the receptive field for each 3D point, thereby effectively preserving geometric details. Extensive experiments show that our RandLA-Net can process 1 million points in a single pass with up to 200x faster than existing approaches. Moreover, our RandLA-Net clearly surpasses state-of-the-art approaches for semantic segmentation on two large-scale benchmarks Semantic3D and SemanticKITTI.

Learning-based visual navigation still remains a challenging problem in robotics, with two overarching issues: how to transfer the learnt policy to unseen scenarios, and how to deploy the system on real robots. In this paper, we propose a deep neural network based visual navigation system, SnapNav. Unlike map-based navigation or Visual-Teach-and-Repeat (VT&R), SnapNav only receives a few snapshots of the environment combined with directional guidance to allow it to execute the navigation task. Additionally, SnapNav can be easily deployed on real robots due to a two-level hierarchy: a high level commander that provides directional commands and a low level controller that provides real-time control and obstacle avoidance. This also allows us to effectively use simulated and real data to train the different layers of the hierarchy, facilitating robust control. Extensive experimental results show that SnapNav achieves a highly autonomous navigation ability compared to baseline models, enabling sparse, map-less navigation in previously unseen environments.

Heart rate monitoring at home is a useful metric for assessing health e.g. of the elderly or patients in post-operative recovery. Although non-contact heart rate monitoring has been widely explored, typically using a static, wall-mounted device, measurements are limited to a single room and sensitive to user orientation and position. In this work, we propose mBeats, a robot mounted millimeter wave (mmWave) radar system that provide periodic heart rate measurements under different user poses, without interfering in a users daily activities. mBeats contains a mmWave servoing module that adaptively adjusts the sensor angle to the best reflection pro le. Furthermore, mBeats features a deep neural network predictor, which can estimate heart rate from the lower leg and additionally provides estimation uncertainty. Through extensive experiments, we demonstrate accurate and robust operation of mBeats in a range of scenarios. We believe by integrating mobility and adaptability, mBeats can empower many down-stream healthcare applications at home, such as palliative care, post-operative rehabilitation and telemedicine.

Inertial information processing plays a pivotal role in egomotion awareness for mobile agents, as inertial measurements are entirely egocentric and not environment dependent. However, they are affected greatly by changes in sensor placement/orientation or motion dynamics, and it is infeasible to collect labelled data from every domain. To overcome the challenges of domain adaptation on long sensory sequences, we propose MotionTransformer - a novel framework that extracts domain-invariant features of raw sequences from arbitrary domains, and transforms to new domains without any paired data. Through the experiments, we demonstrate that it is able to efficiently and effectively convert the raw sequence from a new unlabelled target domain into an accurate inertial trajectory, benefiting from the motion knowledge transferred from the labelled source domain. We also conduct real-world experiments to show our framework can reconstruct physically meaningful trajectories from raw IMU measurements obtained with a standard mobile phone in various attachments.

Modelling the physical properties of everyday objects is a fundamental prerequisite for autonomous robots. We present a novel generative adversarial network (DEFO-NET), able to predict body deformations under external forces from a single RGB-D image. The network is based on an invertible conditional Generative Adversarial Network (IcGAN) and is trained on a collection of different objects of interest generated by a physical finite element model simulator. Defo-netinherits the generalisation properties of GANs. This means that the network is able to reconstruct the whole 3-D appearance of the object given a single depth view of the object and to generalise to unseen object configurations. Contrary to traditional finite element methods, our approach is fast enough to be used in real-time applications. We apply the network to the problem of safe and fast navigation of mobile robots carrying payloads over different obstacles and floor materials. Experimental results in real scenarios show how a robot equipped with an RGB-D camera can use the network to predict terrain deformations under different payload configurations and use this to avoid unsafe areas.

Deep Reinforcement Learning (DRL) has been applied successfully to many robotic applications. However, the large number of trials needed for training is a key issue. Most of existing techniques developed to improve training efficiency (e.g. imitation) target on general tasks rather than being tailored for robot applications, which have their specific context to benefit from. We propose a novel framework, Assisted Reinforcement Learning, where a classical controller (e.g. a PID controller) is used as an alternative, switchable policy to speed up training of DRL for local planning and navigation problems. The core idea is that the simple control law allows the robot to rapidly learn sensible primitives, like driving in a straight line, instead of random exploration. As the actor network becomes more advanced, it can then take over to perform more complex actions, like obstacle avoidance. Eventually, the simple controller can be discarded entirely. We show that not only does this technique train faster, it also is less sensitive to the structure of the DRL network and consistently outperforms a standard Deep Deterministic Policy Gradient network. We demonstrate the results in both simulation and real-world experiments.

Various sensing modalities have been exploited for indoor location sensing, each of which has well understood limitations, however. This paper presents a first systematic study on using the electromagnetic field (EMF) induced by a building's electric power network for simultaneous localization and mapping (SLAM). A basis of this work is a measurement study showing that the power network EMF sensed by either a customized sensor or smartphone's microphone as a side-channel sensor is spatially distinct and temporally stable. Based on this, we design a SLAM approach that can reliably detect loop closures based on EMF sensing results. With the EMF feature map constructed by SLAM, we also design an efficient online localization scheme for resource-constrained mobiles. Evaluation in three indoor spaces shows that the power network EMF is a promising modality for location sensing on mobile devices, which is able to run in real time and achieve sub-meter accuracy.

In pose graph Simultaneous Localization and Mapping (SLAM) systems, incorrect loop closures can seriously hinder optimizers from converging to correct solutions, significantly degrading both localization accuracy and map consistency. Therefore, it is crucial to enhance their robustness in the presence of numerous false-positive loop closures. Existing approaches tend to fail when working with very unreliable front-end systems, where the majority of inferred loop closures are incorrect. In this paper, we propose a novel middle layer, seamlessly embedded between front and back ends, to boost the robustness of the whole SLAM system. The main contributions of this paper are two-fold: 1) the proposed middle layer offers a new mechanism to reliably detect and remove false-positive loop closures, even if they form the overwhelming majority; 2) artificial loop closures are automatically reconstructed and injected into pose graphs in the framework of an Extended Rauch-Tung-Striebel smoother, reinforcing reliable loop closures. The proposed algorithm alters the graph generated by the front-end and can then be optimized by any back-end system. Extensive experiments are conducted to demonstrate significantly improved accuracy and robustness compared with state-of-the-art methods and various back-ends, verifying the effectiveness of the proposed algorithm.