Course ProjectsThoracic Disease Classification and Report GenerationImplemented by: Python, PyTorch. [Project Report] Applied Densenet-121 as image-encoder for multi-label thoracic disease classication on ChestXray-14 dataset and achieved state-of-the-art performance with average AUROC score of 0.83. Implemented weighted loss and oversampling to handle unbalanced data. Built an image-attended 2-layer LSTM decoder for report generation with scaled dot-product attention. Obtained BLEU-1 score of 0.42 and ROUGE score of 0.33. Chinese QA SystemImplemented by: Python. Construct a system which can automatically reply to the natural language questions given by the users. The system can work both online and offline. During the offline mode, the system will find the answer from the local wiki corpus. While during the online mode, the system will use web crawlers to find the answers from the internet. The system include sentence segmentation, keyword extraction, question classification, paragraph search, template matching and some other modules. RISC-V ISA SimulatorImplemented by: C++. [Source Code] Developed a simulator in C++ to simulate the execution of an ELF file based on RISC-V instruction set architecture. Construct relative data structure to simulate the hardware like registers, memory, program counter and so on. The simulator can simulate the executing steps like analyzing, fetching, decoding for a given ELF RISC-V file and also implement most of the system calls like standard input and output. AQI DetectorImplemented by: Swift3, Python. [Source Code] AQI Detector is an interesting IOS application that can help the user to know the real-time AQI at the user's location. User can simply takes a picture to the sky using IOS devices and then upload the picture to the server. At the server side, the relative program will process the picture and uses a SVM algorithm to differentiate the sky area from the rest part and then infer the air quality index(AQI) from the RGB values of the sky area by a pre-trained linear regression model. After finishing the calculation, the server will return the result to the user and the user can know the AQI on the device's screen. Automatic classification of the aviation safety report documentsImplemented by: Python. For a given aviation safety report document, identify the types of flight safety issues involved in the report. Developed in python, using SVM to implement the classification. The precision is over 85 percent. English word reciting softwareImplemented by: Java. Developed in Java and SQL. Include new word reciting, mistakes correcting, stage test, game and some other modules. Embedded a 30000 words offline dictionary and also support inquiring rare words through network. Mine SweepingImplemented by: C++. Implement graphic interface by using Include archive, timing, cheating and some other functions. |