Machine learning could help predict defects, fraud and cancer | Rowan Today

When Dr. Shen-Shyang Ho appears to be like at a graph, he sees greater than summary information factors. In the dynamic graphs he research, the pc science researcher sees complicated networks that change over time. 
An affiliate professor within the Department of Computer Science in Rowan University’s College of Science & Mathematics, Ho has studied and developed machine-learning applied sciences for detecting anomalies in varied utility domains for practically 20 years. 
“Anomalies are deviations from the traditional,” defined Ho, who additionally coordinates Rowan’s grasp’s diploma program in pc science. Many anomalies are undesirable, he added, comparable to monetary fraud, suspicious habits, manufacturing defects and irregular findings on medical assessments. 
Previously, Ho developed novel approaches for figuring out significant anomalies from pictures containing defects in turbine blades or cancerous cells. With a brand new $273,047 grant from the National Science Foundation, he’ll spend the subsequent three years specializing in growing new approaches for detecting, explaining and predicting anomalies in dynamic graphs that repeatedly evolve. 
“When you detect an anomaly, the subsequent stage is decision-making,” Ho stated. “In dynamic graphs, the sooner you may detect or predict anomalies, the higher, as a result of you can begin mitigation efforts. If the anomaly isn’t detected till later, you may not have time.” 
Anomaly detection by machine learning provides alternatives to cease bank card fraud, present early cancer remedies or appropriate manufacturing defects. Deviations in an plane flight sample, for instance, would possibly point out {that a} airplane is dealing with a disaster, comparable to a hijacking. 
Discovering the fast unfold of an infectious virus throughout networks of worldwide airports by detecting anomalies in dynamic graphs representing cross-country human actions permits consultants to implement insurance policies to comprise the unfold of illness.
Dynamic graphs representing an influence grid with substations, transformers and houses can help utility firms perceive and detect irregular utilization patterns, to allow them to higher advise customers on precautions they will take to forestall an outage. 
The advantages of data-driven approaches to detect and predict anomalies have fostered an curiosity in machine learning applied sciences throughout quite a few industries. Early detection reduces monetary prices, whereas preemptively figuring out public well being issues, illness development, manufacturing defects and plane crises can save lives. 
Existing approaches to machine learning supply basic options for anomaly detection, however “good options are problem-specific,” Ho stated. “Our method of particular issues from a number of views extracted from the dynamic graph information permits us to elucidate why we make the decision that an anomaly goes to happen within the close to future.”
“There’s at all times a tradeoff between false positives and false negatives,” Ho added. “If a false optimistic triggers an alarm in your anomaly detection system, your mitigation efforts are losing assets. On the opposite hand, a false detrimental misses the detection of an anomaly which could have catastrophic penalties. Our proposed method permits one to decide on the quantity of false positives he/she will tolerate earlier than the predictive mannequin is even used.”  
“In truth, based mostly on our proposed method,” Ho stated, “I can let you know whether or not there might be an anomaly within the close to future, decide how doubtless it’ll occur, and clarify why I’m making that prediction.”

Recommended For You