• Nuclear Engineering and Technology
• /
• v.55 no.10
• /
• pp.3515-3524
• /
• 2023

A reinforcement learning framework is proposed for the control problem of outlet steam pressure of the once-through steam generator(OTSG) in this paper. The double-layer controller using Proximal Policy Optimization(PPO) algorithm is applied in the control structure of the OTSG. The PPO algorithm can train the neural networks continuously according to the process of interaction with the environment and then the trained controller can realize better control for the OTSG. Meanwhile, reinforcement learning has the characteristic of difficult application in real-world objects, this paper proposes an innovative pretraining method to solve this problem. The difficulty in the application of reinforcement learning lies in training. The optimal strategy of each step is summed up through trial and error, and the training cost is very high. In this paper, the LSTM model is adopted as the training environment for pretraining, which saves training time and improves efficiency. The experimental results show that this method can realize the self-adjustment of control parameters under various working conditions, and the control effect has the advantages of small overshoot, fast stabilization speed, and strong adaptive ability.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.8
• /
• pp.3168-3186
• /
• 2020

Automatic topic essay generation (TEG) is a controllable text generation task that aims to generate informative, diverse, and topic-consistent essays based on multiple topics. To make the generated essays of high quality, a reasonable method should consider both diversity and topic-consistency. Another essential issue is the intrinsic link of the topics, which contributes to making the essays closely surround the semantics of provided topics. However, it remains challenging for TEG to fill the semantic gap between source topic words and target output, and a more powerful model is needed to capture the semantics of given topics. To this end, we propose a pretraining-based contextual self-attention (PC-SAN) model that is built upon the seq2seq framework. For the encoder of our model, we employ a dynamic weight sum of layers from BERT to fully utilize the semantics of topics, which is of great help to fill the gap and improve the quality of the generated essays. In the decoding phase, we also transform the target-side contextual history information into the query layers to alleviate the lack of context in typical self-attention networks (SANs). Experimental results on large-scale paragraph-level Chinese corpora verify that our model is capable of generating diverse, topic-consistent text and essentially makes improvements as compare to strong baselines. Furthermore, extensive analysis validates the effectiveness of contextual embeddings from BERT and contextual history information in SANs.

• ETRI Journal
• /
• v.41 no.2
• /
• pp.242-253
• /
• 2019

We devise a layer-wise hint training method to improve the existing hint-based knowledge distillation (KD) training approach, which is employed for knowledge transfer in a teacher-student framework using a residual network (ResNet). To achieve this objective, the proposed method first iteratively trains the student ResNet and incrementally employs hint-based information extracted from the pretrained teacher ResNet containing several hint and guided layers. Next, typical softening factor-based KD training is performed using the previously estimated hint-based information. We compare the recognition accuracy of the proposed approach with that of KD training without hints, hint-based KD training, and ResNet-based layer-wise pretraining using reliable datasets, including CIFAR-10, CIFAR-100, and MNIST. When using the selected multiple hint-based information items and their layer-wise transfer in the proposed method, the trained student ResNet more accurately reflects the pretrained teacher ResNet's rich information than the baseline training methods, for all the benchmark datasets we consider in this study.

• Annual Conference of KIPS
• /
• 2022.11a
• /
• pp.493-494
• /
• 2022

The development of livestock faces intensive farming results in a rising need for recognition of individual animals such as cows and pigs is related to high traceability. In this paper, we present a non-invasive biometrics systematic approach based on the deep-learning classification model to pig face identification. Firstly, in our systematic method, we build a ROS data collection system block to collect 10 pig face data images. Secondly, we proposed a preprocessing block in that we utilize the SSIM method to filter some images of collected images that have high similarity. Thirdly, we employ the improved image classification model of CNN (ViT), which uses the finetuning and pretraining technique to recognize the individual pig face. Finally, our proposed method achieves the accuracy about 98.66%.

• Journal of Korea Water Resources Association
• /
• v.57 no.7
• /
• pp.437-449
• /
• 2024

Rainfall-runoff prediction studies using deep learning while considering catchment attributes have been gaining attention. In this study, we selected two models: the Transformer model, which is suitable for large-scale data training through the self-attention mechanism, and the LSTM-based multi-state-vector sequence-to-sequence (LSTM-MSV-S2S) model with an encoder-decoder structure. These models were constructed to incorporate catchment attributes and predict the inflow of 10 multi-purpose dam watersheds in South Korea. The experimental design consisted of three training methods: Single-basin Training (ST), Pretraining (PT), and Pretraining-Finetuning (PT-FT). The input data for the models included 10 selected watershed attributes along with meteorological data. The inflow prediction performance was compared based on the training methods. The results showed that the Transformer model outperformed the LSTM-MSV-S2S model when using the PT and PT-FT methods, with the PT-FT method yielding the highest performance. The LSTM-MSV-S2S model showed better performance than the Transformer when using the ST method; however, it showed lower performance when using the PT and PT-FT methods. Additionally, the embedding layer activation vectors and raw catchment attributes were used to cluster watersheds and analyze whether the models learned the similarities between them. The Transformer model demonstrated improved performance among watersheds with similar activation vectors, proving that utilizing information from other pre-trained watersheds enhances the prediction performance. This study compared the suitable models and training methods for each multi-purpose dam and highlighted the necessity of constructing deep learning models using PT and PT-FT methods for domestic watersheds. Furthermore, the results confirmed that the Transformer model outperforms the LSTM-MSV-S2S model when applying PT and PT-FT methods.

• Annual Conference of KIPS
• /
• 2021.05a
• /
• pp.408-410
• /
• 2021

In the NLP field, the pre-training model BERT launched by the Google team in 2018 has shown amazing results in various tasks in the NLP field. Subsequently, many variant models have been derived based on the original BERT, such as RoBERTa, ERNIEBERT and so on. In this paper, the WWMBERT (Whole Word Masking BERT) model suitable for Chinese text tasks was used as the baseline model of our experiment. The experiment is mainly for "Text-level Chinese text classification tasks" are improved, which mainly combines Tapt (Task-Adaptive Pretraining) and "Multi-Sample Dropout method" to improve the model, and compare the experimental results, experimental data sets and model scoring standards Both are consistent with the official WWMBERT model using Accuracy as the scoring standard. The official WWMBERT model uses the maximum and average values of multiple experimental results as the experimental scores. The development set was 97.70% (97.50%) on the "text-level Chinese text classification task". and 97.70% (97.50%) of the test set. After comparing the results of the experiments in this paper, the development set increased by 0.35% (0.5%) and the test set increased by 0.31% (0.48%). The original baseline model has been significantly improved.

• International Journal of Computer Science & Network Security
• /
• v.23 no.8
• /
• pp.177-189
• /
• 2023

Malware detection is an increasingly important operational focus in cyber security, particularly given the fast pace of such threats (e.g., new malware variants introduced every day). There has been great interest in exploring the use of machine learning techniques in automating and enhancing the effectiveness of malware detection and analysis. In this paper, we present a deep recurrent neural network solution as a stacked Long Short-Term Memory (LSTM) with a pre-training as a regularization method to avoid random network initialization. In our proposal, we use global and short dependencies of the inputs. With pre-training, we avoid random initialization and are able to improve the accuracy and robustness of malware threat hunting. The proposed method speeds up the convergence (in comparison to stacked LSTM) by reducing the length of malware OpCode or bytecode sequences. Hence, the complexity of our final method is reduced. This leads to better accuracy, higher Mattews Correlation Coefficients (MCC), and Area Under the Curve (AUC) in comparison to a standard LSTM with similar detection time. Our proposed method can be applied in real-time malware threat hunting, particularly for safety critical systems such as eHealth or Internet of Military of Things where poor convergence of the model could lead to catastrophic consequences. We evaluate the effectiveness of our proposed method on Windows, Ransomware, Internet of Things (IoT), and Android malware datasets using both static and dynamic analysis. For the IoT malware detection, we also present a comparative summary of the performance on an IoT-specific dataset of our proposed method and the standard stacked LSTM method. More specifically, of our proposed method achieves an accuracy of 99.1% in detecting IoT malware samples, with AUC of 0.985, and MCC of 0.95; thus, outperforming standard LSTM based methods in these key metrics.