Tjong:A transformer‐based Mahjong AI via hierarchical decision‐making and fan backward

Mahjong,a complex game with hidden information and sparse rewards,poses significant challenges.Existing Mahjong AIs require substantial hardware resources and extensive datasets to enhance AI capabilities.The authors propose a transformer‐based Mahjong AI(Tjong)via hierarchical decision‐making.By utilising self‐attention mechanisms,Tjong effectively captures tile patterns and game dynamics,and it decouples the decision pro-cess into two distinct stages:action decision and tile decision.This design reduces de-cision complexity considerably.Additionally,a fan backward technique is proposed to address the sparse rewards by allocating reversed rewards for actions based on winning hands.Tjong consists of 15M parameters and is trained using approximately 0.5 M data over 7 days of supervised learning on a single server with 2 GPUs.The action decision achieved an accuracy of 94.63%,while the claim decision attained 98.55%and the discard decision reached 81.51%.In a tournament format,Tjong outperformed AIs(CNN,MLP,RNN,ResNet,VIT),achieving scores up to 230%higher than its opponents.Further-more,after 3 days of reinforcement learning training,it ranked within the top 1%on the leaderboard on the Botzone platform.

Milk-derived exosomes as a promising vehicle for oral delivery of hydrophilic biomacromolecule drugs

Exosomes,as promising vehicles,have been widely used in the research of oral drug delivery,but the generally low drug loading efficiency of exosomes seriously limits its application and transformation.In this study,we systematically investigated the effects of drug loading methods and physicochemical properties(lipophilicity and molecular weight)on drug loading efficiency of milk-derived exosomes to explore the most appropriate loading conditions.Our finding revealed that the drug loading efficiency of exosomes was closely related to the drug loading method,drug lipophilicity,drug molecular weight and exosome/drug proportions.Of note,we demonstrated the universality that hydrophilic biomacromolecule drugs were the most appropriate loading drugs for milk-derived exosomes,which was attributed to the efficient loading capacity and sustained release behavior.Furthermore,milk-derived exosomes could significantly improve the transepithelial transport and oral bioavailability of model hydrophilic biomacromolecule drugs(octreotide,exendin-4 and salmon calcitonin).Collectively,our results suggested that the encapsulation of hydrophilic biomacromolecule drugs might be the most promising direction for milk exosomes as oral drug delivery vehicles.

Boosting low-temperature selective catalytic reduction of NO_(x)with NH_(3)of V_(2)O_(5)/TiO_(2)catalyst via B-doping

A series of B-doped V_(2)O_(5)/TiO_(2) catalysts has been prepared the by sol-gel and impregnation methods to investigate the influence of B-doping on the selective catalytic reduction(SCR)of NOxwith NH_(3).X-ray diffraction,Brunauer-Emmett-Teller specific surface area,scanning electron microscope,X-ray photoelectron spectroscopy,temperature-programmed reduction of H_(2) and temperature-programmed desorption of NH_(3)technology were used to study the effect of the B-doping on the structure and NH_(3)-SCR activity of V_(2)O_(5)/TiO_(2) catalysts.The experimental results demonstrated that the introduction of B not only improved the low-temperature SCR activity of the catalysts,but also broadened the activity temperature window.The best SCR activity in the entire test temperature range is obtained for VTiB_(2.0) with 2.0%doping amount of B and the NO_(x) conversion rate is up to 94.3%at 210℃.The crystal phase,specific surface area,valence state reducibility and surface acidity of the active components for the as-prepared catalysts are significantly affected by the B-doping,resulting in an improved NH_(3)-SCR performance.These results suggest that the V_(2)O_(5)/TiO_(2) catalysts with an appropriate B content afford good candidates for SCR in the low temperature window.

Tibetan Multi-Dialect Speech Recognition Using Latent Regression Bayesian Network and End-To-End Mode

We proposed a method using latent regression Bayesian network (LRBN) toextract the shared speech feature for the input of end-to-end speech recognition model.The structure of LRBN is compact and its parameter learning is fast. Compared withConvolutional Neural Network, it has a simpler and understood structure and lessparameters to learn. Experimental results show that the advantage of hybridLRBN/Bidirectional Long Short-Term Memory-Connectionist Temporal Classificationarchitecture for Tibetan multi-dialect speech recognition, and demonstrate the LRBN ishelpful to differentiate among multiple language speech sets.

Cross-Language Transfer Learning-based Lhasa-Tibetan Speech Recognition

As one of Chinese minority languages,Tibetan speech recognition technology was not researched upon as extensively as Chinese and English were until recently.This,along with the relatively small Tibetan corpus,has resulted in an unsatisfying performance of Tibetan speech recognition based on an end-to-end model.This paper aims to achieve an accurate Tibetan speech recognition using a small amount of Tibetan training data.We demonstrate effective methods of Tibetan end-to-end speech recognition via cross-language transfer learning from three aspects:modeling unit selection,transfer learning method,and source language selection.Experimental results show that the Chinese-Tibetan multi-language learning method using multilanguage character set as the modeling unit yields the best performance on Tibetan Character Error Rate(CER)at 27.3%,which is reduced by 26.1%compared to the language-specific model.And our method also achieves the 2.2%higher accuracy using less amount of data compared with the method using Tibetan multi-dialect transfer learning under the same model structure and data set.

TibetanGoTinyNet:a lightweight U-Net style network for zero learning of Tibetan Go

The game of Tibetan Go faces the scarcity of expert knowledge and research literature.Therefore,we study the zero learning model of Tibetan Go under limited computing power resources and propose a novel scaleinvariant U-Net style two-headed output lightweight network TibetanGoTinyNet.The lightweight convolutional neural networks and capsule structure are applied to the encoder and decoder of TibetanGoTinyNet to reduce computational burden and achieve better feature extraction results.Several autonomous self-attention mechanisms are integrated into TibetanGoTinyNet to capture the Tibetan Go board’s spatial and global information and select important channels.The training data are generated entirely from self-play games.TibetanGoTinyNet achieves 62%–78%winning rate against other four U-Net style models including Res-UNet,Res-UNet Attention,Ghost-UNet,and Ghost Capsule-UNet.It also achieves 75%winning rate in the ablation experiments on the attention mechanism with embedded positional information.The model saves about 33%of the training time with 45%–50%winning rate for different Monte–Carlo tree search(MCTS)simulation counts when migrated from 9×9 to 11×11 boards.Code for our model is available at https://github.com/paulzyy/TibetanGoTinyNet.

Combination Mechanism and Enhanced Visible-Light Photocatalytic Activity and Stability of CdS/g-C3N4 Heterojunctions

In this study, CdS/g-C3N4 （CSCN） heterojunctions were in situ fabricated with a large amount of CdS nanoparticles anchored on g-C3N4 nanosheets, A wet chemical method was developed for the first time to determine the actual content of CdS in CSCN composites. X-ray diffraction （XRD）, Fourier transform infrared spectra （FFIR）, high-resolution transmission electron microscopy （HRTEM） and UV-vis diffuse reflectance spectra （DRS） were employed to characterize the composition, structure and optical prop- erty of CSCN composites. Based on the is0electric point （liP） analysis of g-C3N4, a conclusion was obtained on the combination mechanism between CdS nanoparticles and g-C3N4 nanosheets. The photocatalytic activity of CSCN composites was much better than those of individual CdS and g-C3N4 for the degrada- tion of azo dye Methyl Orange （MO） by 40 min adsorption in the dark followed by 15 min photocatalysis under visible light irradiation. After 5 cycles, CSCN composites still maintained high reactive activity with the MO degradation efficiency of 93.8%, exhibiting good photocatalytic stability. The Cd2~ concentration dissolved in the supernatant detected by atomic absorption spectroscopy （AAS） of CSCN composites was lower than that of pure CdS, implying that the photocorrosion of CdS could be suppressed via the combination with g-C3N4. Photoluminescence emission spectra （PL） results clearly revealed that the recombination of photogenerated electron-hole pairs in CSCN composites was effectively inhibited due to the formation of heterojunctions. Based on the band alignments of g-C3N4 and CdS, the possible photocatalvtic mechnism was discussed.

A modified YOLOv4 detection method for a vision-based underwater garbage cleaning robot

To tackle the problem of aquatic environment pollution,a vision-based autonomous underwater garbage cleaning robot has been developed in our laboratory.We propose a garbage detection method based on a modified YOLOv4,allowing high-speed and high-precision object detection.Specifically,the YOLOv4 algorithm is chosen as a basic neural network framework to perform object detection.With the purpose of further improvement on the detection accuracy,YOLOv4 is transformed into a four-scale detection method.To improve the detection speed,model pruning is applied to the new model.By virtue of the improved detection methods,the robot can collect garbage autonomously.The detection speed is up to 66.67 frames/s with a mean average precision(mAP)of 95.099%,and experimental results demonstrate that both the detection speed and the accuracy of the improved YOLOv4 are excellent.