When you ask Alexa have a question – say to buy more peanut butter – you probably have a specific idea in mind. As if playing a game of 20 questions, Alexa is trying to meet your request in as few turns as possible and pinpoining the context and specific.
Last summer, as an intern at the Amazon Alexa AI team in Sunnyvale, California, Qing Guo worked on the project to help Alexa understand the user’s intention. The project drew on Guo’s background as a Ph.D. -Students in statistics at Virginia Tech: She used statistical techniques to improve education and performance for machine learning models that power Alexa.
For example, she used statistical concepts, such as weight weight and variation conference for contact -life learning, which allowed the model to focus only on the most relevant answers during the model education process. This makes model training more stable and efficient. Her work is a group in information theory that provides a principle of measurement and quantification of information on information contained in question-answers PEIRS.
“By incorporating these statistical techniques, I was able to improve the performance of the algorithm,” Guo said. “The algorithm could be trained using small batch sizes without compromising the Ovall performance, making it available and effective for real-time interaction. This is important for training large models, such as deep neural nets.”
Now Guo is back at Amazon for another internship attached to a scholarship awarded to Virginia Tech Doctorate Students through the Amazon -Virginia Tech Initiative for Effective and Robust Machine Learning.
The initiative launched in March 2022 is focused on research on effective and robust machine learning. Among other things, it provides an opportunity for doctorate students who perform AI and ML research to apply for Amazon fellowships.
Guo applied for the community for the opportunity to try his academic ideas in the real world, industrial surroundings. Her submission was accepted and her subsequent interaction with Amazon scientists, she said, provides “insight and inspiration from different perspectives to continuously strengthen my research.”
E.g. The interactions focused her research on solutions to problems that are simple, robust replacements for components of models used in the real world applications.
An introduction to machine learning
Guo studied applied statistics at Shanghai University of International Business and Economics in China. She didn’t know much about machine learning until she came to Virginia Tech, where her adviser, Xinwei Deng, asked her to code a statistical solution for a machine learning problem using Python.
Guo taught Python and began collaborating with computer science students and academic colleagues, drawing on methods that use theories and concepts from statistics to improve training and performance for machine learning models.
More broadly, Guo said that her Ph.D. -Research center on strategies to extract the most valuable information from data while using calculation resources more effectively.
“Improving data quality or more effectively extracting information from data has been increasingly important for machine learning, and that’s what the lawnutans are good at,” she said. “This is formally known as experience design in the statistical literature, which is less known for the machine learning community and has great potential to improve machine learning practices.”
E.g. At Virginia Tech, she and Chenyang Tao, a used scientist who is now hon mentor at Amazon, are the world together about the development of a technique that enables the use of small data sets to train machine learning models for computer vision and natural language processing. Exercise models for these types of applications typically require large data sets and abuant computer resources.
Work gearing statistical concepts such as mutual information that measures the dependence between variables and variation conference, as Guo said, “is a powerful tool because it is reformulated complex, expensive problem with simple, cheap, statistical accuracy.”
Their approach is eight times more effective than the current advanced solution she nadud. The researchers presented their work at the conference on neural information treatment system (Neurips) in 2022, and since then they have continued to improve the technique, which is fandamental for Guo’s Ph.D. -research.
Use of Statistics for Machine Learning at Amazon
During the first internship, Guo had weekly meetings with colleagues at Alexa AI, who helped her apply her statistical skills to the real world’s machine learning problems. For example, they advised that she was down to a strategy to understand customer efforts not only from incomplete, but also wrong information, such as the wrong actor’s name when searching for a movie.
“I have to think about this problem and improve my model,” Guo said. “This is a very valuable insight. Today, when I do my research, I always think, ‘Is there anything else I need to think about?'”
For his second internship, Guo Tao and his team about basic research for a next generation machine learning technique will help customize computer vision and language models to enable multimodal models to answer questions with multimedia information.
She is also involved in investigative interviews with internal teams about new ways of training large language models with limited, targeted data, reducing the training time of generative AI systems.
Guo believes that training and fine-tuning of generative AI models with only the most informative data will overcome some of the computer resource restrictions to train these types of models.
“This is a very hot topic now,” Guo said. “Multimodal is a very important area to investigate. I am very lucky to be assigned this project.”
A dedication to academia
Tao Notice the industry experience that Guo got during her internships will inform her academic research in the future.
“She wants to know firsthand the challenges that used researchers in the industry are facing,” he said. “This will give her a lot of opportunities for new research topics and a direct path to her research to influence people’s lives.
In the end, Guo hopes to pursue a career in Academy with his Ph.D. in statistics.
“My dream is to be a professor,” she said. “I think research is very interesting as I can solve different kinds of problems that I am interested in.”
There will be another lesson from Interning at Amazon to prove the key: the need to communicate with colleagues and teammates approaching projects from different perspectives and areas of expertise, Guo noted.
In her weeks’ meetings with Mentors and leaders at Amazon, she has learned how communication skills advanced their careers. To be a professor of academia, it is important to use “simple words to express technical concepts to people with different backgrounds who want to solve the problem,” she said.