Of all the sketchy stuff out of Queen’s recently this could take the biscuit. Queen’s University is helping Israel develop “deep learning” on the activities of Palestinians.
Deep learning is on the horizon for Northern Ireland, it is in part already here. We have a number of Facial recognition systems companies itching to institute it.
As we mentioned in a previous article Queen’s University has been working on preparing the next generation of digital authoritarians. One Prof. Neil Robertson is an organiser for the 13th Annual IEEE (The Institute of Electrical and Electronics Engineers) on automatic face and gesture recognition. The other main organisers are from Israel and China. Two arguably despotic regimes.
In this short article Robertson appeals for ideas improving big data. Basically there is already a huge amount of data being gathered on individuals, the Government and companies will be able to track you around Belfast, they will know if you are sad, happy, angry, they will know what you posted on social media etc, etc.
Mr Robertson previously worked for the MOD on a number of Military applications including hot targeting software. So this guy is not a pacifist.
What Dr Robertson might see if he looked at himself with his own deep data scanner.
From his article:
In recent years, Deep Learning (DL) has become a dominant method for a wide variety of computer vision tasks. One of its biggest successes has been in face recognition where the performance has been improved dramatically. So, will DL make other face recognition algorithms obsolete? Is deep learning always the best solution in any scenarios? Is it necessary for researchers to deeply investigate the traditional face recognition technique in the DL era? Actually, deep learning is not perfect. For instance, deep learning heavily depends on big data which is sometimes quite expensive and sometimes may not be available. Due to this limitation, conventional methods achieve superior or comparable performance against DL methods in the field of facial landmark detection, face recognition across large poses, thermal/near Infrared Face Recognition, 3D face recognition, etc. It would be interesting to explicitly compare DL methods with traditional methods in terms of accuracy, efficiency and model complexity. We aim to investigate the scenarios where conventional methods can outperform DL methods by EXPLICIT comparison and deep analysis.
We welcome submissions on topics related to the invesigation of (i) the advantages of traditional methods against DL methods and (ii) insights into the reasons of these advantages. Submissions will be peer-reviewed and should follow the standard FG2018 IEEE format.
This is hugely iffy.