Competition on robust deep learning.
Ontology highlight
ABSTRACT:
SUBMITTER: Dong Y
PROVIDER: S-EPMC10257479 | biostudies-literature | 2023 Jun
REPOSITORIES: biostudies-literature
Publications
Competition on robust deep learning.
Dong Yinpeng Y Liu Chang C Xiang Wenzhao W Su Hang H Zhu Jun J
National science review 20230407 6
This perspective paper proposes a new adversarial training method based on large-scale pre-trained models to achieve state-of-the-art adversarial robustness on ImageNet. ...[more]
Similar Datasets
Project description:Evolution of cooperation and competition can appear when multiple adaptive agents share a biological, social, or technological niche. In the present work we study how cooperation and competition emerge between autonomous agents that learn by reinforcement while using only their raw visual input as the state representation. In particular, we extend the Deep Q-Learning framework to multiagent environments to investigate the interaction between two learning agents in the well-known video game Pong. By manipulating the classical rewarding scheme of Pong we show how competitive and collaborative behaviors emerge. We also describe the progression from competitive to collaborative behavior when the incentive to cooperate is increased. Finally we show how learning by playing against another adaptive agent, instead of against a hard-wired algorithm, results in more robust strategies. The present work shows that Deep Q-Networks can become a useful tool for studying decentralized learning of multiagent systems coping with high-dimensional environments.
| S-EPMC5381785 | biostudies-literature
Project description:The automatic detection and recognition of sound events by computers is a requirement for a number of emerging sensing and human computer interaction technologies. Recent advances in this field have been achieved by machine learning classifiers working in conjunction with time-frequency feature representations. This combination has achieved excellent accuracy for classification of discrete sounds. The ability to recognise sounds under real-world noisy conditions, called robust sound event classification, is an especially challenging task that has attracted recent research attention. Another aspect of real-word conditions is the classification of continuous, occluded or overlapping sounds, rather than classification of short isolated sound recordings. This paper addresses the classification of noise-corrupted, occluded, overlapped, continuous sound recordings. It first proposes a standard evaluation task for such sounds based upon a common existing method for evaluating isolated sound classification. It then benchmarks several high performing isolated sound classifiers to operate with continuous sound data by incorporating an energy-based event detection front end. Results are reported for each tested system using the new task, to provide the first analysis of their performance for continuous sound event detection. In addition it proposes and evaluates a novel Bayesian-inspired front end for the segmentation and detection of continuous sound recordings prior to classification.
| S-EPMC5593179 | biostudies-other
Project description:Recently, modern smartphones equipped with a variety of embedded-sensors, such as accelerometers and gyroscopes, have been used as an alternative platform for human activity recognition (HAR), since they are cost-effective, unobtrusive and they facilitate real-time applications. However, the majority of the related works have proposed a position-dependent HAR, i.e., the target subject has to fix the smartphone in a pre-defined position. Few studies have tackled the problem of position-independent HAR. They have tackled the problem either using handcrafted features that are less influenced by the position of the smartphone or by building a position-aware HAR. The performance of these studies still needs more improvement to produce a reliable smartphone-based HAR. Thus, in this paper, we propose a deep convolution neural network model that provides a robust position-independent HAR system. We build and evaluate the performance of the proposed model using the RealWorld HAR public dataset. We find that our deep learning proposed model increases the overall performance compared to the state-of-the-art traditional machine learning method from 84% to 88% for position-independent HAR. In addition, the position detection performance of our model improves superiorly from 89% to 98%. Finally, the recognition time of the proposed model is evaluated in order to validate the applicability of the model for real-time applications.
| S-EPMC6263408 | biostudies-other
Project description:Although deep learning has revolutionized protein structure prediction, almost all experimentally characterized de novo protein designs have been generated using physically based approaches such as Rosetta. Here, we describe a deep learning-based protein sequence design method, ProteinMPNN, that has outstanding performance in both in silico and experimental tests. On native protein backbones, ProteinMPNN has a sequence recovery of 52.4% compared with 32.9% for Rosetta. The amino acid sequence at different positions can be coupled between single or multiple chains, enabling application to a wide range of current protein design challenges. We demonstrate the broad utility and high accuracy of ProteinMPNN using x-ray crystallography, cryo-electron microscopy, and functional studies by rescuing previously failed designs, which were made using Rosetta or AlphaFold, of protein monomers, cyclic homo-oligomers, tetrahedral nanoparticles, and target-binding proteins.
| S-EPMC9997061 | biostudies-literature
Project description:Recent critical commentaries unfavorably compare deep learning (DL) with standard machine learning (SML) approaches for brain imaging data analysis. However, their conclusions are often based on pre-engineered features depriving DL of its main advantage - representation learning. We conduct a large-scale systematic comparison profiled in multiple classification and regression tasks on structural MRI images and show the importance of representation learning for DL. Results show that if trained following prevalent DL practices, DL methods have the potential to scale particularly well and substantially improve compared to SML methods, while also presenting a lower asymptotic complexity in relative computational time, despite being more complex. We also demonstrate that DL embeddings span comprehensible task-specific projection spectra and that DL consistently localizes task-discriminative brain biomarkers. Our findings highlight the presence of nonlinearities in neuroimaging data that DL can exploit to generate superior task-discriminative representations for characterizing the human brain.
| S-EPMC7806588 | biostudies-literature
Project description:The correct estimation of gait events is essential for the interpretation and calculation of 3D gait analysis (3DGA) data. Depending on the severity of the underlying pathology and the availability of force plates, gait events can be set either manually by trained clinicians or detected by automated event detection algorithms. The downside of manually estimated events is the tedious and time-intensive work which leads to subjective assessments. For automated event detection algorithms, the drawback is, that there is no standardized method available. Algorithms show varying robustness and accuracy on different pathologies and are often dependent on setup or pathology-specific thresholds. In this paper, we aim at closing this gap by introducing a novel deep learning-based gait event detection algorithm called IntellEvent, which shows to be accurate and robust across multiple pathologies. For this study, we utilized a retrospective clinical 3DGA dataset of 1211 patients with four different pathologies (malrotation deformities of the lower limbs, club foot, infantile cerebral palsy (ICP), and ICP with only drop foot characteristics) and 61 healthy controls. We propose a recurrent neural network architecture based on long-short term memory (LSTM) and trained it with 3D position and velocity information to predict initial contact (IC) and foot off (FO) events. We compared IntellEvent to a state-of-the-art heuristic approach and a machine learning method called DeepEvent. IntellEvent outperforms both methods and detects IC events on average within 5.4 ms and FO events within 11.3 ms with a detection rate of ≥ 99% and ≥ 95%, respectively. Our investigation on generalizability across laboratories suggests that models trained on data from a different laboratory need to be applied with care due to setup variations or differences in capturing frequencies.
| S-EPMC10420363 | biostudies-literature
Project description:Accurate prediction of enzyme kinetic parameters is crucial for enzyme exploration and modification. Existing models face the problem of either low accuracy or poor generalization ability due to overfitting. In this work, we first developed unbiased datasets to evaluate the actual performance of these methods and proposed a deep learning model, CataPro, based on pre-trained models and molecular fingerprints to predict turnover number (kcat), Michaelis constant (Km), and catalytic efficiency (kcat/Km). Compared with previous baseline models, CataPro demonstrates clearly enhanced accuracy and generalization ability on the unbiased datasets. In a representational enzyme mining project, by combining CataPro with traditional methods, we identified an enzyme (SsCSO) with 19.53 times increased activity compared to the initial enzyme (CSO2) and then successfully engineered it to improve its activity by 3.34 times. This reveals the high potential of CataPro as an effective tool for future enzyme discovery and modification.
| S-EPMC11923063 | biostudies-literature
Project description:Interest in autonomous vehicles (AVs) is growing at a rapid pace due to increased convenience, safety benefits and potential environmental gains. Although several leading AV companies predicted that AVs would be on the road by 2020, they are still limited to relatively small-scale trials. The ability to know their precise location on the map is a challenging prerequisite for safe and reliable AVs due to sensor imperfections under adverse environmental and weather conditions, posing a formidable obstacle to their widespread use. Here we propose a deep learning-based self-supervised approach for ego-motion estimation that is a robust and complementary localization solution under inclement weather conditions. The proposed approach is a geometry-aware method that attentively fuses the rich representation capability of visual sensors and the weather-immune features provided by radars using an attention-based learning technique. Our method predicts reliability masks for the sensor measurements, eliminating the deficiencies in the multimodal data. In various experiments we demonstrate the robust all-weather performance and effective cross-domain generalizability under harsh weather conditions such as rain, fog and snow, as well as day and night conditions. Furthermore, we employ a game-theoretic approach to analyse the interpretability of the model predictions, illustrating the independent and uncorrelated failure modes of the multimodal system. We anticipate our work will bring AVs one step closer to safe and reliable all-weather autonomous driving.
| S-EPMC10543073 | biostudies-literature
Project description:BackgroundT-cells play a crucial role in the adaptive immune system by triggering responses against cancer cells and pathogens, while maintaining tolerance against self-antigens, which has sparked interest in the development of various T-cell-focused immunotherapies. However, the identification of antigens recognised by T-cells is low-throughput and laborious. To overcome some of these limitations, computational methods for predicting CD8 + T-cell epitopes have emerged. Despite recent developments, most immunogenicity algorithms struggle to learn features of peptide immunogenicity from small datasets, suffer from HLA bias and are unable to reliably predict pathology-specific CD8 + T-cell epitopes.MethodsWe developed TRAP (T-cell recognition potential of HLA-I presented peptides), a robust deep learning workflow for predicting CD8 + T-cell epitopes from MHC-I presented pathogenic and self-peptides. TRAP uses transfer learning, deep learning architecture and MHC binding information to make context-specific predictions of CD8 + T-cell epitopes. TRAP also detects low-confidence predictions for peptides that differ significantly from those in the training datasets to abstain from making incorrect predictions. To estimate the immunogenicity of pathogenic peptides with low-confidence predictions, we further developed a novel metric, RSAT (relative similarity to autoantigens and tumour-associated antigens), as a complementary to 'dissimilarity to self' from cancer studies.ResultsTRAP was used to identify epitopes from glioblastoma patients as well as SARS-CoV-2 peptides, and it outperformed other algorithms in both cancer and pathogenic settings. TRAP was especially effective at extracting immunogenicity-associated properties from restricted data of emerging pathogens and translating them onto related species, as well as minimising the loss of likely epitopes in imbalanced datasets. We also demonstrated that the novel metric termed RSAT was able to estimate immunogenic of pathogenic peptides of various lengths and species. TRAP implementation is available at: https://github.com/ChloeHJ/TRAP .ConclusionsThis study presents a novel computational workflow for accurately predicting CD8 + T-cell epitopes to foster a better understanding of antigen-specific T-cell response and the development of effective clinical therapeutics.
| S-EPMC10498576 | biostudies-literature
Project description:The bubble point pressure (P b) is a crucial pressure-volume-temperature (PVT) property and a primary input needed for performing many petroleum engineering calculations, such as reservoir simulation. The industrial practice of determining P b is by direct measurement from PVT tests or prediction using empirical correlations. The main problems encountered with the published empirical correlations are their lack of accuracy and the noncomprehensive data set used to develop the model. In addition, most of the published correlations have not proven the relationships between the inputs and outputs as part of the validation process (i.e., no trend analysis was conducted). Nowadays, deep learning techniques such as long short-term memory (LSTM) networks have begun to replace the empirical correlations as they generate high accuracy. This study, therefore, presents a robust LSTM-based model for predicting P b using a global data set of 760 collected data points from different fields worldwide to build the model. The developed model was then validated by applying trend analysis to ensure that the model follows the correct relationships between the inputs and outputs and performing statistical analysis after comparing the most published correlations. The robustness and accuracy of the model have been verified by performing various statistical analyses and using additional data that was not part of the data set used to develop the model. The trend analysis results have proven that the proposed LSTM-based model follows the correct relationships, indicating the model's reliability. Furthermore, the statistical analysis results have shown that the lowest average absolute percent relative error (AAPRE) is 8.422% and the highest correlation coefficient is 0.99. These values are much better than those given by the most accurate models in the literature.
| S-EPMC8388097 | biostudies-literature