41530 research outputs found
Sort by
Improving Document Classification by Integrating Human-Crafted Semantic Knowledge
Get PDFDocument classification is a pivotal task in various domains, warranting the development of robust algorithms. Among these, the Bidirectional Encoder Representations from Transformers (BERT) algorithm, introduced by Google, has proven to perform well when fine-tuned for the task at hand. Leveraging transformer architecture, BERT demonstrates stellar language understanding capabilities. However, the integration of BERT with a range of approaches has shown potential for further enhancing classification accuracy. This paper investigates three techniques that leverage semantic understanding to improve the performance of document classification models trained with BERT. First, we balance corpora afflicted by imbalanced training data distributions. Next, we substitute words and phrases with semantically similar terms to create “synthetic documents,” thereby shifting the focus from individual words to semantic meaning. Finally, we retrain our model on a dataset composed of “synthetic documents” with heavier weights given to classes with commonly misclassified documents during the initial round of training. These approaches emphasize the significance of semantic comprehension in document classification, as the meaning of words and phrases often relies on contextual cues. Our findings demonstrate the efficacy of these techniques and highlight the importance of incorporating semantic knowledge in document classification algorithms. On our labeled testing set of news articles from BBC News, BERT performs with a baseline macro F1 score of 0.902. Incorporating the three techniques, we were able to improve the macro F1 score to 0.951
Perceptual Hash Based Content Matching
Get PDFThe proliferation of video content and AI generated imagery has introduced a number of new challenges in content identification and verification. The ability to trace content back to its source has become a critical problem as video content increases both naturally and synthetically through AI generation. This thesis provides the design, analysis, and experimental verification for a perceptual hash based framework aimed at addressing these challenges. Perceptual hashing is a method for encoding the visual content of images into compact and easily comparable binary strings. This process is used as the foundation for content matching in videos and source verification in AI generated content.
The proposed content matching process is evaluated through two primary applications. The process is first applied to video content in which individual frames are matched across a dataset of car crash compilations. This experiment demonstrates the effectiveness of the perceptual hash in locating identical frames that may have minor visual transformations. The process is then applied to AI generated images, where its ability to identify the source image is tested. Through varying denoising parameters during the AI image generation process, the method’s sensitivity to subtle changes in image content can be assessed. The results of this thesis demonstrate the advantages and disadvantages of perceptual hashing in both application
The Integration and Validation of a Class 3B Laser for Stereo Particle Imaging Velocimetry
Get PDFParticle Imaging Velocimetry (PIV) is a laser flow diagnostic measurement technique used to measure spatial and temporal velocity fields. In an effort to promote experimental aerodynamic research, specifically an entry point for students to learn Particle Imaging Velocimetry, an in-house Class 3B laser PIV system was developed and its experimental results were verified with a Class 4 laser. Both systems were integrated on the Cal Poly Water Tunnel Laboratory. The work emphasizes the application of lasers controlled by a micro-controller to perform microsecond-based pulses and integrating a newly-developed device with an old system consisting of software, cameras, and synchronizers. The flow measurements in both freestream and two specific cylinder cases were found to demonstrate consistency between the systems. The Class 3B laser system opens an avenue for making the PIV technique more accessible for laboratory classes while significantly reducing the risks surrounding laser safety that are typically common with Class 4 laser systems. A collection of freestream and cylinder cases were used to compare the performance of the system to the Class 4 Laser and demonstrated that the Class 3B laser system is a viable approach for flow with low Reynolds number flows and velocities up to 6.8 cm/s
Rover Chassis & Suspension
Get PDFThe Mars rover project is part of the University Rover Challenge (URC), a robotics competition organized by the Mars Society that tasks student teams with designing and building the next generation of Mars rovers. Cal Poly Human Space Technology and Research (CP-HSTAR) has inherited a rover developed by a previous senior design team.
This report documents the progress made in refining the rover’s structure, highlighting key specifications, and detailing design improvements aimed at enhancing the suspension’s structural integrity, improving wheel traction, increasing modularity, and developing an enclosed chassis for electronic components.
This report includes a detailed breakdown of the rover’s chassis, differential, and suspension design. As well as the implementation of the verification prototype, design verification results, and future work required to optimize the prototype. The appendices provide supporting documentation such as the user manual, risk assessment, budget, and test procedures
Using Phosphorus-Deprived, Filamentous Microalgae to Remove Soluble Phosphorus from Tertiary Municipal Wastewater
No full textPhosphorus (P) is a nutrient that is essential for crops, but it is a non-renewable resource. P recovery from wastewater would lessen P pollution and extend the P supply for fertilizers. Filamentous microalgae can remove soluble inorganic P from water and assimilate it into recyclable biomass. To further develop this concept by using P-depleted filamentous microalgae, this research pursued three goals: to determine (1) the biomass-specific P uptake rates of Tribonema minus and Uronema sp., (2) how long Uronema sp. can be cultivated in P-depleted state (not P-starved) and continue substantial uptake, and (3) if the P dosing rates impact the uptake response and/or productivity of Uronema sp. Raceway tanks were given little or no soluble P to generate P-depleted biomass. The P-depleted biomass was then used for uptake contact experiments in which P uptake rates and biomass P content were measured. The long duration (0-10 h) uptake rates were not substantially different for T. minus and Uronema sp., but Uronema sp. tended to uptake more quickly in the short duration (0-2 and 0-3 h) of the contact period. Other experiments focused on prolonged deprivation, during which the raceways received P every day or every three days, although the mass of P dosed over the long term was equivalent in all raceways. Uronema sp. could be cultivated in a P-depleted state for an average of 10 days before the biomass was unable to have substantial P uptake. The uptake rates for these two dosing regimens were assumed to be the same because the 0-6 hour average rate was 0.33 mg P/g VSS-h (dosed every day) and 0.36 mg P/g VSS-h (dosed every three days). Future studies should confirm if Uronema sp. consistently assimilates more P at a faster rate in the winter compared to the spring, as observed in the present study
