Cyclic peptides, known for their high binding affinity and low toxicity, show potential as innovative drugs targeting "undruggable" proteins. However, therapeutic efficacy is often hindered by poor membrane permeability. Over the past decade, FDA has approved an average of one macrocyclic peptide drug per year, with romidepsin being only intracellular site. Biological experiments to measure permeability are time-consuming labor-intensive. Rapid assessment cyclic crucial development. In this...

The goal of text-to-image synthesis is to generate a visually realistic image that matches given text description. In practice, the captions annotated by humans for same have large variance in terms contents and choice words. linguistic discrepancy between identical leads synthetic images deviating from ground truth. To address this issue, we propose contrastive learning approach improve quality enhance semantic consistency images. pretraining stage, utilize learn consistent textual...

Diffusion Denoising Probability Models (DDPM) and Vision Transformer (ViT) have demonstrated significant progress in generative tasks discriminative tasks, respectively, thus far these models largely been developed their own domains. In this paper, we establish a direct connection between DDPM ViT by integrating the architecture into DDPM, introduce new model called Generative (GenViT). The modeling flexibility of enables us to further extend GenViT hybrid discriminative-generative modeling,...

In this paper, the time series classification frontier method MiniRocket was used to classify earthquakes, blasts, and background noise. From supervised unsupervised classification, a comprehensive analysis carried out, finally, achieved excellent results. The relatively simple model, MiniRocket, is only one-dimensional convolutional neural network structure which has best results, its computational efficiency far stronger than other methods. Through our experimental we found that model...

Abstract Despite the widespread use of ionizable lipid nanoparticles (LNPs) in clinical applications for messenger RNA (mRNA) delivery, mRNA drug delivery system faces an efficient challenge screening LNPs. Traditional methods often require a substantial amount experimental time and incur high research development costs. To accelerate early stage LNPs, we propose TransLNP, transformer-based transfection prediction model designed to aid selection LNPs systems. TransLNP uses two types...

Wee1 is a kinase that regulates cell cycle arrest in response to DNA damage. inhibition potential strategy suppress the growth of tumors with defective p53 or repair pathways. However, development inhibitors faces some challenges. AZD1775, first-in-class inhibitor, has poor selectivity and dose-limiting toxicity. Here, we report discovery

Joint Energy-based Model (JEM) [12] is a recently proposed hybrid model that retains strong discriminative power of modern CNN classifiers, while generating samples rivaling the quality GAN-based approaches. In this paper, we propose variety new training procedures and architecture features to improve JEM's accuracy, stability, speed altogether. 1) We proximal SGLD generate in proximity from previous step, which improves stability. 2) further treat approximate maximum likelihood learning EBM...

Can we train a hybrid discriminative-generative model with single network? This question has recently been answered in the affirmative, introducing field of Joint Energy-based Model (JEM) [17], [48], which achieves high classification accuracy and image generation quality simultaneously. Despite recent advances, there remain two performance gaps: gap to standard softmax classifier, state-of-the-art generative models. In this paper, introduce variety training techniques bridge JEM. 1) We...

Processing large point clouds is a challenging task. Therefore, the data often downsampled to smaller size such that it can be stored, transmitted and processed more efficiently without incurring significant performance degradation. Traditional task-agnostic sampling methods, as farthest (FPS), do not consider downstream tasks when clouds, thus non-informative points are sampled. This paper explores task-oriented for 3D aims sample subset of tailored specifically task interest. Similar FPS,...

The success of Deep Neural Networks (DNNs) highly depends on data quality. Moreover, predictive uncertainty makes high performing DNNs risky for real-world deployment. In this paper, we aim to address these two issues by proposing a unified filtering framework leveraging underlying density, that can effectively denoise training as well avoid predicting uncertain test points. Our proposed leverages distribution differentiate between noise and clean samples without requiring any modification...

The demand for object pose estimation is steadily increasing, and deep learning has propelled the advancement of this field. However, majority research endeavors face challenges in their applicability to industrial production. This primarily due high cost annotating 3D data, which places higher demands on generalization capabilities neural network models. Additionally, existing methods struggle handle abundance textureless objects commonly found settings. Finally, there a strong real-time...

As the primary mRNA delivery vehicles, ionizable lipid nanoparticles (LNPs) exhibit excellent safety, high transfection efficiency, and strong immune response induction. However, screening process for LNPs is time-consuming costly. To expedite identification of high-transfection-efficiency drug systems, we propose an explainable efficiency prediction model, called TransMA. TransMA employs a multi-modal molecular structure fusion architecture, wherein fine-grained atomic spatial relationship...

Adversarial Training (AT) and Virtual (VAT) are the regularization techniques that train Deep Neural Networks (DNNs) with adversarial examples generated by adding small but worst-case perturbations to input examples. In this paper, we propose xAT xVAT, new training algorithms generate multiplicative for robust of DNNs. Such much more perceptible interpretable than their additive counterparts exploited AT VAT. Furthermore, can be transductively or inductively, while standard VAT only support...

Knowledge graphs typically contain a large number of entities but often cover only fraction all relations between them (i.e., incompleteness). Zero-shot link prediction (ZSLP) is popular way to tackle the problem by automatically identifying unobserved entities. Most recent approaches use textual features (e.g., surface name or descriptions) as auxiliary information improve encoded representation. These methods lack robustness they are bound support tokens from fixed vocabulary and unable...

Adversarial Training (AT) and Virtual (VAT) are the regularization techniques that train Deep Neural Networks (DNNs) with adversarial examples generated by adding small but worst-case perturbations to input examples. In this paper, we propose xAT xVAT, new training algorithms, generate \textbf{multiplicative} for robust of DNNs. Such much more perceptible interpretable than their \textbf{additive} counterparts exploited AT VAT. Furthermore, multiplicative can be transductively or inductively...

The success of Deep Neural Networks (DNNs) highly depends on data quality. Moreover, predictive uncertainty reduces reliability DNNs for real-world applications. In this paper, we aim to address these two issues by proposing a unified filtering framework leveraging underlying density, that effectively denoises training as well avoids predicting confusing samples. Our proposed differentiates noise from clean samples without modifying existing DNN architectures or loss functions. Extensive...

Energy-based models (EBMs) exhibit a variety of desirable properties in predictive tasks, such as generality, simplicity and compositionality. However, training EBMs on high-dimensional datasets remains unstable expensive. In this paper, we present Manifold EBM (M-EBM) to boost the overall performance unconditional Joint Model (JEM). Despite its simplicity, M-EBM significantly improves stability speed host benchmark datasets, CIFAR10, CIFAR100, CelebA-HQ, ImageNet 32x32. Once class labels...

This paper proposes a transformer fault diagnosis method based on the fusion of Dissolved Gas Analysis (DGA) information. The aims to address issues uncertainty, redundancy, and singularity in traditional methods for transformers. By utilizing Principle Component (PCA) technique, 16 gas ratio features are effectively reduced dimensionality. dimensionally feature parameters then utilized as inputs Extreme Learning Machines (ELM), Backpropagation (BP) neural networks, Support Vector (SVM)....

The method of dissolved gas analysis (DGA) in oil is one the important methods for transformer fault detection. However, due to shortcomings traditional diagnostic methods, such as incomplete coding and too absolute coding, they can not meet actual diagnosis needs. In recent years, intelligent algorithms have been widely used. Aiming at Extreme Learning Machine (ELM), poor training stability low accuracy, this paper proposes a through Hunter Prey Optimizer (HPO) optimize ELM. At same time,...

The existing transformer fault diagnosis methods have the problems of single information source and low diagnostic accuracy, making it difficult to make accurate comprehensive judgments on actual situation transformers. On basis multi-source fusion in power transformers, this paper proposes an improved method that combines HPO-ELM D-S evidence theory. optimize output through mapping obtain probability outputs for different labels, then use theory fuse allocation matrix. were compared...

Due to the incompleteness of knowledge graphs (KGs), zero-shot link prediction (ZSLP) which aims predict unobserved relations in KGs has attracted recent interest from researchers. A common solution is use textual features (e.g., surface name or descriptions) as auxiliary information bridge gap between seen and unseen relations. Current approaches learn an embedding for each word token text. These methods lack robustness they suffer out-of-vocabulary (OOV) problem. Meanwhile, models built on...

Teacher racial diversity has been widely considered important in education. However, it remains unclear to what extent and how teacher addressed at the federal, state, district levels. In this study, we employed text mining collect analyze over three million documents We found that while students of color had disproportionately less access racially diverse teachers, under our analysis insufficiently discussed recruitment retention teachers. Our findings also reveal education agencies levels...

Can we train a hybrid discriminative-generative model within single network? This question has recently been answered in the affirmative, introducing field of Joint Energy-based Model (JEM), which achieves high classification accuracy and image generation quality simultaneously. Despite recent advances, there remain two performance gaps: gap to standard softmax classifier, state-of-the-art generative models. In this paper, introduce variety training techniques bridge JEM. 1) We incorporate...

Joint Energy-based Model (JEM) of Grathwohl et al. shows that a standard softmax classifier can be reinterpreted as an energy-based model (EBM) for the joint distribution p(x,y); resulting optimized to improve calibration, robustness, and out-of-distribution detection, while generating samples rivaling quality recent GAN-based approaches. However, JEM exploits is inherently discriminative its latent feature space not well formulated probabilistic distributions, which may hinder potential...