LONDON, Sept. 19, 2018 (GLOBE NEWSWIRE) -- One of the top data mining conferences, ACM SIGKDD (Knowledge Discovery and Data Mining Conference, abbreviated as KDD), was held from August 19 to 23rd in London, U.K. this year. The annual KDD conference is a premier interdisciplinary data science conference, bringing together researchers and practitioners from data science, data mining, knowledge discovery, and large-scale data analytics. It also attracts the world's leading technology companies to attend and present including Google, Facebook, and Microsoft along with Chinese Technology giants Alibaba, Tencent, Baidu and JD.
As the only Chinese Company in the AI Adaptive learning field, many attendees were very interested to learn much more about how Yixue Squirrel AI worked—specifically, the AI and learning systems that it has recently developed along with plans for new systems that will be coming soon. Moreover, Dr. Dan Bindman, Chief Data Scientist of Squirrel AI, gave a talk on “Multi- Dimensional Models for Knowledge Assessment and AI Adaptive Learning: Now and the Future”. He described the competitive advantages of Squirrel AI in efficiently and accurately assessing student knowledge of the curriculum, pinpointing exactly each student’s strengths and weaknesses at the highest level of granularity and providing personalized teaching. He also discussed the key AI algorithms of Squirrel AI, the current experiments his team has been developing and their focus on R&D in the future. These developments seemed very well received and quite impressed many professors, scientists, and industry practitioners attending the conference.
A photo accompanying this announcement is available at http://www.globenewswire.com/NewsRoom/AttachmentNg/ec2b5921-eafb-4338-99b9-afa57dc7cf5e
In addition, Squirrel AI also announced further expansions of its open data plan. It is looking to use its already vast student data in collaborative projects with global researchers and scientists to deliver the absolute state of the art AI adaptive learning experience.
Great attendees, great talks, great panels:
Some of the important attendees discussing their research and the frontiers of artificial intelligence and machine learning are given below:
- Usama M. Fayyad, former Chief Data Officer at Barclays Bank and co-founder of KDD conferences and ACM SIGKDD
- Jeannette M. Wing, Avanessians Director of the Data Sciences Institute at Columbia University, former Corporate Vice President of Microsoft Research with oversight of its core research laboratories around the world and Microsoft Research Connections, and former President's Professor of Computer Science at Carnegie Mellon University
- Yike Guo, Professor of Computing Science at Imperial College London, Academician of the Royal Academy of Engineering, and the founding Director of the Data Science Institute at Imperial College
- Alvin E. Roth, Nobel laureate in economics and professor of economics at Stanford University
- Joseph Sirosh, Microsoft AI Chief Technology Officer
- Yee Whye, Professor of machine learning at Oxford University
- Bo Xing, Machine Learning Professor at Carnegie Mellon Lung
- Andrej KARPATHY, Tesla AI Director
- Faisal Farooq, IBM chief scientist
A photo accompanying this announcement is available at http://www.globenewswire.com/NewsRoom/AttachmentNg/5468ed6e-0cc2-4508-9714-97cb3b556beb
After learning about the cutting-edge algorithms and the results that Squirrel AI has achieved, Yike Guo and Jeannette Wing, along with key AI practitioners from Facebook and other leading technology companies, were among those who seemed impressed by what Squirrel AI has achieved in successfully applying AI in education, as well as their commitment to bringing equitable, efficient and personalized education to every Chinese child.
Also of note, Yee Whye The, Professor of Statistical Machine Learning at the University of Oxford, was one of the keynote speakers at KDD 2018. He gave a speech of On Big Data Learning for Small Data Problems. As of June 14th, 2018, his papers have been cited 20,142 times! Prior to working at the University of Oxford, Yee Whye Teh worked as a University Lecturer at the Department of Statistics University College London. He was one of the original developers of deep belief networks and of hierarchical dirichlet processes. He has also served on many occasions as chairperson on NIPS, ICML and AISTATS.
Dr. Dan Bindman, Chief Data Scientist of Squirrel AI, gives a powerful talk on AI adaptive learning:
Dan’s talk first focused on the 3 Key Components of an “Ultimate” AI Adaptive Learning System:
- The first component is “complete and deep content”, meaning (a) complete curriculum coverage using high-quality “nuts and bolts” questions, with all the ancillaries like engaging videos and explanations, at a high level of granularity for easier student learning, and (b) a large and diverse set of more complex exercises that gives students a much deeper understanding of the curriculum. These exercises are often open-ended and thus difficult to create and score for student understanding. An expert in adaptive learning content, Dan believes that Yixue Squirrel AI is already very good in providing strong “nuts and bolts” content coverage across our wide variety of K-12 products. The extent of the deeper content, however, is currently limited by business needs, as many users want deep content only if it substantially improves their success on the high stakes tests that they are studying for. But in the next 5 to 10 years, Dan believes providing a strong deep content system will become the “make or break” for the success of many AI adaptive learning systems, and he has been very impressed how Yixue Squirrel AI has been investing heavily in this area to be sure it is at the forefront of AI adaptive learning.
- The second component is a strong student knowledge assessment system, which can efficiently and accurately assess student knowledge of the entire curriculum, exactly pinpointing each student’s strengths and weaknesses at the highest level of granularity. And ideally it can be updated in real time using the student’s history on the product, without the requirement of formal periodic assessments or tests except for a short initial assessment to more accurately initialize the system. Again, Dan was impressed by the systems currently used by Squirrel AI in this regard, including Item Response Theory, Knowledge Space Theory, and Bayesian Knowledge Tracing. And perhaps most especially impressive is how some of the subjects that Squirrel AI is covering, like Ancient Chinese, have never been implemented with adaptive AI technology until now! And going forward, he is excited by how Squirrel AI is willing to try multi dimensional, mixed model approaches, along with using exciting new models (like his own, see below) to achieve the most accurate assessments. And he is hopeful that within the next year Squirrel AI will be able to update the student’s knowledge of the entire course without periodic assessments.
- The third and final component is a content recommendation system that maximizes student learning. In the classical sense, the primary input to such a system is the student’s current assessed knowledge. The system then uses that to recommend one (or more) lessons or questions that are at just the right difficulty for the student to learn. But Dan also thinks it is becoming clear that these system must focus more on student engagement—adjusting the recommendation depending on the emotional state of the student. Almost certainly there are times that the student needs easier problems to get out of a rut. There may also be other times when a reward, like a game, will be a great motivator. Or there may be other times when a challenging problem will be best. And again, Dan was impressed by Squirrel AI’s current, classical recommendation system, but he is even more impressed by how Squirrel AI is working and investing in researching new approaches to maximize student engagement—like real-time heart rate, brain wave, and facial recognition analysis along with a Virtual Personal Assistant to get an “emotional measure” on the student that will lead to better recommendations. He also discussed how Squirrel AI is currently working with the Stanford Research Institute (SRI) and the Chinese Academy of Sciences on many of these things and is just about to begin to collect some larger-scale student data.
In the second part of his talk, Dan talked about new model that he is working with Squirrel AI to implement. The key in his model is Probabilistic Knowledge State, or PKS. The PKS for any given student on any given question at any given time is determined by three factors: The student’s abilities A(t), the question’s weight W(q), and the question’s center C(q). After the student answers only 15-20 questions the model can accurately estimate the PKS for the student on every question, which tells us the student’s strengths and weaknesses at the highest level of granularity. And because the PKS is so accurate and is updated continuously as the student answers more questions, it is very useful for both student knowledge analytics and AI adaptive learning—basically providing an “X-Ray” of what the student’s current knowledge is and at the same time providing for an optimal way to choose the next question or lesson for the student.
Finally, Dan also shared some other of Squirrel AI R&D’s current achievements and future plan. All attendees displayed a strong interest in Dan’s model, Squirrel AI’s algorithms and the current projects.
More about KDD and China’s growing influence at the conference:
KDD was established in 1998 by the Association for Computing Machinery, or ACM. KDD has hosted the annual ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) since 1995, and conference papers are published through ACM. KDD is widely considered the most influential forum for knowledge discovery and data mining research. Each year, over 2000 global top data-mining scholars as well as representatives from leading companies from all over the world, attend KDD. This year, 293 papers were accepted at KDD, and 64.5% or 189 of these accepted papers had at least one Chinese author.
About YiXue Squirrel AI Learning Inc.
YiXue Squirrel AI Inc. is a leading AI-based adaptive learning service provider for K-12 students in China. Headquartered in Shanghai, China, YiXue offers after-school courses for Math, English, Chinese, Physics and Chemistry subjects, powered by its proprietary AI adaptive engine and custom-built courseware. Students on Squirrel AI 's platform enjoy a supervised adaptive learning experience that has been proven to improve both efficacy and student engagement across Squirrel AI's online learning platform and in-person learning centers.
To learn more about YiXue Squirrel AI Learning, please visit http://www.squirrelai.us/.