ProfessorOmid Kavehei

Project Opportunities

Title: Low-power Intelligent Bio-Signal Processing

Summary of opportunity:

 This project aims at developing a responsive implant that is making decisions based on smart bio-signal processing.

Opportunity synopsis:

 While artificial intelligence (AI) has paved its way to some bionic applications mostly through software post-processing, close to none of today's electronic implants can be named "Intelligent"! We believe with the large amount of data that is available in any medical field and remarkable advances in AI and the expertise that we have in that as well as electronic technology shrinkage to a remarkable scale, there is a real chance that we may ultimately be able to let an implant to "learn and understand" streams of neural data and make decision with high accuracy. There are software and hardware expertise exist in our side that requires to be matched with medical knowledge, expertise and data. With that in mind please find the following proposal for collaboration [1-4].

This project aims to (A) introduce a deep learning platform for prediction of anomalies before it causes alteration of consciousness or other damages, (B) implement an ultra-low power fully digital and fully customized chipset to parallel achieved software performance, (C) mitigate problems of ‘false alarms' and ‘delay in action' confirmed using a set of clinical trials.

This project uses our state-of-the-art GPU cluster to develop the software and integrated circuit design tools to explore hardware implementation. We also study packaging and high-level integration and surgery issues for full animal clinical trial. 

References:

1. Tran, Nhan, et al. "A complete 256-electrode retinal prosthesis chip." IEEE Journal of Solid-State Circuits 49.3 (2014): 751-765.
2. Truong, Nhan Duy, et al. "Supervised Learning in Automatic Channel Selection for Epileptic Seizure Detection." Expert Systems with Applications (2017)
3. Truong, Nhan Duy, et al. "A Generalised Seizure Prediction with Convolutional Neural Networks for Intracranial and Scalp Electroencephalogram Data Analysis." arXiv preprint arXiv:1707.01976 (2017).
4. Ahnood, Arman, et al. "Retinal Implants: Diamond Devices for High Acuity Prosthetic Vision." Advanced Biosystems 1.1-2 (2017).

Web URL: www.deepnano.ai

How to Apply: To apply, please email omid.kavehei@sydney.edu.au with the subject line "PhD Application”, and attach the following:

• CV
• Transcripts

--------------------------------------------------

Project Opportunities

Title: Real-time Low-power Neural Accelerator

Summary of opportunity:

 This project aims at developing an Application-Specific Integrated Circuit (ASIC) for a low-power learning system processing real-time time-series data.

Opportunity synopsis:

 For over a decade, the semiconductor industry, that propelled us all into the digital age, has been struggling with a chip power-density crisis. The birth of multi-cores microprocessors has its root in this very challenging issue of fundamental physics. The known thermal-voltage (kT/q) parameter, which identifies the scalability of basic transistor parameters, is a non-scalable factor. Additionally, the minimum feature size of the transistors has continuously shrunk from 10μm in 1970 to just 14nm in 2015 (iPhone 6S was shipped with 14nm transistors) making it extremely difficult and costly to engineer energy barriers in transistor channels (where electrons move) to avoid excessive electron tunneling through the barrier when the transistor is supposed to be OFF. Success of the industry has traditionally been measured by how effectively they avoid quantum mechanical effects, such as the tunneling. Unfortunately, with a few tens of atoms across a sub-10nm transistor switch, such effectiveness is vanishing into the shadow of ever increasing quantum tunneling. This project aims to develop a neuro-inspired computing platform for cognitive task that current supercomputers fail to do in real-time like human brain [1-3]. The system architecture is inspired by how the brain function and solely dedicated to tasks dealing with big-data, including data mining, data analytics, pattern recognition and feature extraction.We use our state-of-the-art GPU cluster to develop the software and integrated circuit design tools to explore hardware implementation.

References:
[1] Kavehei, Omid, and Efstratios Skafidas. "Highly scalable neuromorphic hardware with 1-bit stochastic nano-synapses." 2014 IEEE International Symposium on Circuits and Systems (ISCAS).
[2] Kornijcuk, Vladimir, et al. "Multiprotocol-induced plasticity in artificial synapses." Nanoscale 6.24 (2014): 15151-15160.
[3] Kavehei, Omid, Efstratios Skafidas, and Kamran Eshraghian. "Memristive in situ computing." Memristor Networks. Springer International Publishing, 2014. 413-428.

How to Apply: To apply, please email omid.kavehei@sydney.edu.au with the subject line PhD Application,  and attach the following:

• CV
• Transcripts

--------------------------------------------------

Project Opportunities

Title: Data-intensive solutions for medical technologies

Summary of opportunity:

This project aims at developing advanced solutions in software for processing medical data, addressing the issue of less or no labelled data.

Opportunity synopsis:

Using today's advances in machine intelligence and pattern recognition, and our incredibly massive amount of structured and unstructured data on central nervous systems and the Brain, making sense of massive datasets with high amount of data-noise and incoherency is now a possibility [1-3]. We will develop, test and implement cognitive computing technologies in data-driven medical contexts. This project aims to develop data-driven machine learning medical technologies to make medical practices more personalized and precision in both domains of medical devices and services. While expanding knowledge in the information and computing sciences, this project aims to massively reduce costs in health and support services as well as providing low-cost bed-side or wearable technologies for constant monitoring and notification systems. This project uses our state-of-the-art GPU cluster to develop these technologies.

--------------------------------------------------

Project Opportunities

Title: Unobtrusive Sensor Systems for Healthcare

Research Area:

Wearable sensors, unobtrusive sensing, embedded systems, health monitoring.

Opportunity synopsis:

This project aims to create unobtrusive sensor systems to monitor physiological and behavioural data of human body. The signals include EEG, ECG, EOG, EMG, respiration, SpO2, temperature, activities, etc.. The candidate will develop innovative wearable sensing or non-contact sensing techniques with new sensing materials and embedded systems for biomedical data acquisition and processing. Lightweight AI algorithms and software for embedded sensor systems will be proposed for intelligent healthcare systems. The potential applications will be sleep monitoring, neonatal monitoring and smart rehabilitation, etc.

--------------------------------------------------

Project Opportunities

Title: Data Processing for Sleep Monitoring

Research Area:

Signal processing, AI and deep learning, sleep staging and sleep disease analysis.

Opportunity synopsis:

This project focuses on development of signal processing and AI methods and algorithms for sleep staging and sleep disorder analysis. The scope will include proposing dedicated AI and data science methods to explore and analyse relevant biomarkers and features with high sensitivity and specificity from physiological and behavioural signals; building data analytical models to enhance the learning performance, proposing approaches to improve model generalization; investigating explainable AI methods; developing information and computing schemes and algorithms for precise sleep disease detection and prediction.

--------------------------------------------------

Project Opportunities

Title: Real-time Low-power Neural Accelerator

Summary of opportunity:

This project aims at developing an Application-Specific Integrated Circuit (ASIC) for a low-power learning system processing real-time time-series data.

Opportunity synopsis:

For over a decade, the semiconductor industry, that propelled us all into the digital age, has been struggling with a chip power-density crisis. The birth of multi-cores microprocessors has its root in this very challenging issue of fundamental physics. The known thermal-voltage (kT/q) parameter, which identifies the scalability of basic transistor parameters, is a non-scalable factor. Additionally, the minimum feature size of the transistors has continuously shrunk from 10μm in 1970 to just 14nm in 2015 (iPhone 6S was shipped with 14nm transistors) making it extremely difficult and costly to engineer energy barriers in transistor channels (where electrons move) to avoid excessive electron tunnelling through the barrier when the transistor is supposed to be OFF. Success of the industry has traditionally been measured by how effectively they avoid quantum mechanical effects, such as the tunnelling. Unfortunately, with a few tens of atoms across a sub-10nm transistor switch, such effectiveness is vanishing into the shadow of ever increasing quantum tunnelling. This project aims to develop a neuro-inspired computing platform for cognitive task that current supercomputers fail to do in real-time like human brain [1-3]. The system architecture is inspired by how the brain function and solely dedicated to tasks dealing with big-data, including data mining, data analytics, pattern recognition and feature extraction. We use our state-of-the-art GPU cluster to develop the software and integrated circuit design tools to explore hardware implementation.

--------------------------------------------------

Project Opportunities

Title: Data-intensive solutions for medical technologies

Summary of opportunity:

This project aims at developing advanced solutions in software for processing medical data, addressing the issue of less or no labelled data.

Opportunity synopsis:

Using today's advances in machine intelligence and pattern recognition, and our incredibly massive amount of structured and unstructured data on central nervous systems and the Brain, making sense of massive datasets with high amount of data-noise and incoherency is now a possibility [1-3]. We will develop, test and implement cognitive computing technologies in data-driven medical contexts. This project aims to develop data-driven machine learning medical technologies to make medical practices more personalized and precision in both domains of medical devices and services. While expanding knowledge in the information and computing sciences, this project aims to massively reduce costs in health and support services as well as providing low-cost bed-side or wearable technologies for constant monitoring and notification systems. This project uses our state-of-the-art GPU cluster to develop these technologies.

How to Apply: To apply, please email omid.kavehei@sydney.edu.au with the subject line PhD Application and attach the following:

• CV
• Transcripts

References:

[1]. Shen, Dinggang, Guorong Wu, and Heung-Il Suk. "Deep learning in medical image analysis." Annual Review of Biomedical Engineering 0 (2017).[2]. Truong, Nhan Duy, et al. "Supervised Learning in Automatic Channel Selection for Epileptic Seizure Detection." Expert Systems with Applications (2017).[3]. Truong, Nhan Duy, et al. "A Generalised Seizure Prediction with Convolutional Neural Networks for Intracranial and Scalp Electroencephalogram Data Analysis." arXiv preprint arXiv:1707.01976 (2017).