The Open AI Team (OAIT)

RESEARCH PROJECTS

OBJECTIVE

• OPEN means an open platform that supports your success by offering our 30+ year's AI/ML critical expertise & resources to anyone, anywhere, and anytime. Together we make the world better.
• OPEN means collaboration, as it is self-evident that "The whole is greater than the sum of its parts" (Mark Twain: collaborate with a real expert not a fake)
• OPEN means fairness and sharing, and the OAIT supports the open-source ecosystem by offering free access to our papers, code, and data (rob or help?)
• The goal of the OAIT is to advance beyond the LLMs toward building the LIMs, LVMs, LXMs, LMMs, and LOMs:

• Large Image Models (LIMs)
• Large Video Models (LVMs)
• Large X-ray Models (LXMs)
• Large MRI Models (LMMs)
• Large Omics Models (LOMs)

by leveraging our critical expertise in advanced statistical learning, image/video analytics, computer vision, pattern recognition, learning, and AI.

• Example #1, the Bayes classifier yields the minimum classification error, and the OAIT's research on Bayse classifier design has led to the development of the Bayesian Discriminating Features (BDF) method that was awarded a patent on face detection.

• Example #2, the statistical learning theory indicates that the risk functional consists of two terms: the empirical risk and the structure risk defined by the VC dimension. To minimize the overall risk, the OAIT has developed the multiclass Kernel Fisher Analysis (KFA) method, which won the best performance on the government organized large-scale face recognition grand challenge (FRGC) competition. Note that another participating team from CMU proposed a filtering approach that rests on the convolution operation. Convolution was later incorporated into the convolutional neural network or CNN/LeCun that underlies much of the modern deep learning technology (DeepBlue/Chess, DeepQA/Watson, DeepMind/AlphaGo, DeepLearning/ChatGPT, DeepSeek/LLMs...), which inflames the current AI frenzy.

• Myth or Fact
  

• Myth: AI is destroying jobs, as evidenced by Amazon's laying off 14k workers; intimidating Tesla's humanoid Optimus...
• Fact: No, AI is not destroying the job market, but rather it's shifting people from tedious low end to high paying jobs: OAIT attracts 15 talents in 3 weeks (founded on the Mooncake Festival of 2025), and will reach 50 by next year, 500 soon, 1,000 not far away...

• The godfather of AI, Jensen Huang on October 8's CNBC interview commented on his investment in the AI startups: I only regret I did not invest enough! Yeah Bro, does OAIT sound like a 2nd chance? Your comment is so true.

• Jensen on the interview also classified AI into 4 levels (cf. David Marr's 3 levels of complex information-processing systems):

Applications

1. AI/ML for Smart Energy by OAIT
2. AI/ML for Smart Health -- Towards Developing AI Doctor & Digital Assistant by OAIT
3. Video Analytics for Reducing Traffic Congestion, Improving Traffic Safety, and Autonomous Driving by OAIT
4. Facial Recognition for Enhancing Security and Public Safety by OAIT

Models

1. LLMs by OpenAI, DeepSeek
2. LIMs, LVMs, LXMs, LMMs, and LOMs by OAIT

Chips -- kudos for US dominance

1. NVIDIA
2. AMD
3. Intel

Energy

1. AI/ML for Smart Energy by OAIT

↑ Back to Top

MEMBERS

Awards & Publication

[G1].  NJIT Grace Hopper Artificial Intelligence Research Institute seed grant, NJIT, 2025-2026.

1. Z. Yu, K.S. Ganni, J.A. Mehta, P. Pong, J. Li, and C. Liu, "Machine Learning and AI for Optimizing and Safeguarding Energy Transmission in Storms by Automatic Inspection of Electrical Wires", 2025, IEEE Xplore
2. C. Yang, R. Yesgari, V. Vora, P. Pong, J. Li, and C. Liu, "Preventing Wildfires in Energy Transmission by Automatic Power Line Defects Detection Using Machine Learning and AI", 2025, IEEE Xplore
3. C. Yang and C. Liu, "Towards Building AI Doctor and Digital Assistant by Capitalizing on A Mixture of AI Medical Systems", the 6th International Conference on Medical Imaging and Computer-Aided Diagnosis, Nov. 19-21, 2025, London, UK
4. C. Yang, R. Yesgari, and C. Liu, "Privacy-Preserving Automated Rosacea Detection Based on Medically Inspired Region of Interest Selection", IEEE International Conference on Electrical and Computer Engineering Researches, 06-08 December 2025, Antananarivo, MADAGASCAR
5. C. Yang and C. Liu, "Investigating the Impact of Various Loss Functions and Learnable Wiener Filter for Laparoscopic Image Desmoking", the 6th International Conference on Medical Imaging and Computer-Aided Diagnosis, Nov. 19-21, 2025, London, UK
6. C. Yang, R. Yesgari, and C. Liu, "Patch-based Automatic Rosacea Detection Using the ResNet Deep Learning Framework", the 6th International Conference on Medical Imaging and Computer-Aided Diagnosis, Nov. 19-21, 2025, London, UK
7. C. Yang and C. Liu, "Laparoscopic Image Desmoking Using the U-Net with New Loss Function and Integrated Differentiable Wiener Filter", IEEE the 11th International Conference on Big Data Computing Service and Machine Learning Applications, July 21-24, 2025, Tucson, Arizona
8. C. Yang and C. Liu, "Interpretable Automatic Rosacea Detection with Whitened Cosine Similarity", IEEE the 17th International Conference on Computer Research and Development, January 17-19, 2025, Jiangxi, China
9. here for more papers

AREA TOPICS

I. AI/ML for Energy Transition Optimization and Smart Energy on the Wulver/GPU platform

We plan to apply various innovative and advanced AI, deep learning, statistical learning methods for energy transition optimization and smart energy. Specifically, we will apply our advanced AI and deep learning approaches to automatically inspect and protect electrical wires from being damaged by the falling tree limbs during major storms. Such damages not only bring inconvenience to communities, but may also endanger human lives because the falling live wires pose the potential danger of electrocuting people. We will also apply our advanced image and video analysis technologies to inspect the electrical wires automatically for potential risks that may cause wild fires in the woods and remote areas. Some demos and preliminary results using our innovative AI, deep learning, computer vision, and pattern recognition methods are publicly accessible at (https://web.njit.edu/~cliu/AISH.html) for AI Doctor & Digital Assistant for providing the best care for anyone, anywhere, and anytime; at (https://web.njit.edu/~cliu/NJDOT/DEMOS.html) for automated traffic incidents detection and traffic congestion detection for improving traffic safety and autonomous driving; at (https://frvp.njit.edu) for advanced facial detection and recognition for enhancing security and public safety. We will leverage these innovative methods for addressing energy transition optimization and smart energy tasks.

II. AI/ML for Smart Health using the Wulver/GPU platform to Develop AI Doctor & Digital Assistant

The World Health Organization (WHO) estimates that 5.7 to 8.4 million deaths annually are attributable to poor quality care, i.e., people reach care but get substandard or delayed care. Another study reveals that 8.6 million excess annual deaths were amenable to health care of which 5.0 million were due to poor-quality care and 3.6 million lack of health care.

To meet these challenges and save lives, an interdisciplinary team from the NJIT, the Rutgers medical school, and the MSKCC in New York, proposes to move from isolated models to a unified ecosystem of interoperable AI systems that together form an AI doctor & digital assistant platform. This framework will help improve the quality of life & the health of the country by providing the best care for everyone, everywhere, and at any time, and as a result, the excessive waiting time like weeks or months to see a doctor will be completely eliminated. Note that NASA and Google currently are also developing AI Doctor & Digital Assistant for astronauts who are planning a trip to Mars. In contrast to NASA and Google's project, which uses large language models or LLMs, our project applies digital images, digital videos, X-ray, CT scans, MRIs (e.g., brain imaging), Positron Emission Tomography (PET) scans etc. to integrate a mixture of AI medical systems into the AI doctor & digital assistant framework for all people anywhere and anytime.

III. Video Analytics on the Wulver/GPU platform for Reducing Traffic Congestion, Improving Traffic Safety, Home Security, Smart UAS/Drones, and Autonomous Driving

In the United States alone, traffic congestion costs hundreds of billions of dollars annually in direct and indirect losses. Through added emissions of harmful substances, they also contribute to environment degradation and global warming, and adversely affect people's quality of life. It is thus urgent to develop innovative ways to slow or even reverse the trend of growing congestion. This research will integrate advanced traffic detection system, wireless communications, distributed computing, sensing technologies, and cooperative vehicle-infrastructure architecture. It fits well with the Smart City initiative driven by the U.S. Department of Transportation (USDOT) to make our cities more livable, safer, faster, and greener.

Recently FOX Business reported that "Tesla under federal investigation over self-driving cars allegedly breaking traffic laws". For example, Wrong-Way Driving -- "Teslas crossing double-yellow lines, entering oncoming traffic or attempting to turn onto roads in the wrong direction." The OAIT has developed innovative and advanced video analytics technologies to save human lives. Some video analytics demos are able to detect the following traffic incidents:

• TRAJECTORY CONFLICT DETECTION
• STOPPED VEHICLE DETECTION
• WRONG-WAY VEHICLE DETECTION
• SLOW SPEED DETECTION
• CONGESTION DETECTION
• PEDESTRIAN DETECTION
• ACCIDENT DETECTION
• VEHICLE CLASSIFICATION
• VEHICLE COUNTING

IV. Facial Recognition using the Wulver/GPU for Enhancing Security and Public Safety

The OAIT has developed innovative and advanced statistical kernel methods, such as the multiclass Kernel Fisher Analysis (KFA) method and achieved the best performance on the government organized large-scale face recognition grand challenge (FRGC) competition. The novel and advanced statistical methods for facial recognition include a slew of new color models and innovative color feature extraction approaches, the Novel Locally Linear KNN Method, A Sparse Representation Model Using the Complete Marginal Fisher Analysis Framework, new efficient SVM (eSVM), Clustering-based Discriminant Analysis, Feature Local Binary Patterns, New Color SIFT Descriptors, the Bayes Decision Rule Induced Similarity Measures, a novel Bayesian Discriminating Features Method, the enhanced Independent Component Analysis (eICA), the new Gabor Feature based Classifier (GFC), the Evolutionary Pursuit (EP) method, the Probabilistic Reasoning Models (PRMs), and the Enhanced Fisher linear discriminant Models (EFMs).

Q&A

(a break from Boeing's 777 working schedule, meant for relaxing&amusing)

Milestones/Breakthroughs/Contributions

Readings