During a pandemic, patient transfer and ICU bed availability management are life-saving capabilities that can be optimized with smoother communication in many countries.

Today, Hospitals lose precious time reserving ICU beds by phone. It is in the best interest of society that the full efforts of medical institutions remain focused on taking care of patients and not the logistics involved in the process. It is imperative that these logistics of reserving and transferring beds and patients be digitalized and automated. Let us take a case study for France. In France, it is organized at the national level, where a crisis unit within the General Directorate of Health (DGS) decides COVID-19 patient transfers. It has to call the various Regional Health Authorities (ARS) to find out the reception capacities of hospitals, and then turn to ambulances to organize the trips. Logistics are complex because the transfer process involves many actors, and each must take into account the saturation of establishments, which evolves from hour to hour and depends on the condition of the sick.

One of today’s key challenges in the field is to swiftly request bed transfers and try to predict bed occupancies as much as possible in light of COVID-19 patients’ conditions rapidly deteriorating. On the authorities level, there is an urgent need for the centralization of the data nationwide to have an easy at a glance overview of all transfers requests coming from hospitals and their ICU bed availability.

We connect health authorities and hospitals via customized dashboards to save lives during the COVID-19 epidemic.

Our solution revolves around centralizing data on ICU bed availability nationwide, enabling simple and rapid patient transfer on a larger scale. Using machine learning models, based on WHO medical indicators, we can help hospitals predict which COVID-19 patients are more prone to being admitted to ICU over the coming hours. Using feedback from varied healthcare professionals (nurses, ER doctors, and ER professionals) we have been able to develop a simple interface for HCPs to input only essential vitals for patients, providing vital data for the predictive model to perform risk group analysis and ICU flagging. Thanks to Huddle, hospitals will be able to track bed availability, request patient transfers with a simple click, and prepare for incoming patient transfers in advance.

We provide health authorities with a user-friendly dashboard that features: Bed ICU occupancy and availability nationwide Real time patient transfer requests from hospitals System recommendation of hospitals to transfer patients to (based on distance, hospital ICU bed availabilities including a buffer utilizing our predictive patient deterioration model), Live Map.

We provide hospitals (head nurse or head of department) an easy-to-use interface for tracking: Bed ICU occupancy and availability at hospital level Patient health deterioration using non linear machine learning models Patient transfer requests Incoming patients transfers and preparation of their arrival

When uploaded to the cloud, patient names are replaced by a patient ID, and data is limited based on its relevance for the prediction model. We can ensure data protection and privacy standards.

The initial prototype is near completion and the deep learning neural network is getting better day by day as more data becomes available. After the campaign we plan to productionize the app on cloud and partner with a pilot hospital and health authority to validate our solution. Then we aim to open source the entire application with our implementation available to the public health authorities which would be billed per use by hospitals.

• _A novel framework for image cryptography using neural networks, with a shallow encoder and a deep decoder network. Involved both basic and applied research.
• _Encoder is computationally simple, designed for real-time encoding on edge devices.
• _Decoder is a sophisticated network which is run on a secure, computationally powerful machine.
• _Peer review from JVIS (Elsevier) pointed that the research and approach is both interesting and novel, but more sophisticated methodology must be tested. I Intend to pursue this further, and send an updated manuscript for review.

Inspired by a paper on Neural Cryptography presented at NIPS 2018, which showed how deep learning could be used for cryptanalysis and breaking ciphers, I decided to work on an idea of mine that I had conceived a year before, related to encrypting and decrypting images. Pertinent to my interests, I further modified it to specifically target edge devices for real-time encryption. I did extensive research on previous and current work on these topics, read many papers and sections of books, and contacted some professors and doctoral students who I met in conferences or contacted over the years. After the hypothesis and methodology was finalized, and the basic research done so that the framework worked in theory, I implemented it using python, TensorFlow, CUDA, etc. I followed the scientific method to the word and made several variations of both (encryptor and decryptor) the models till the framework worked. After a month of development and experimentation, I was able to get some success with about 50% of the images getting decrypted in a recognizable form from a simple encrypted floating-point vector. I drafted a paper and got the preprint reviewed by my advisors, and then sent it to JVIS (Elsevier) for peer review. While they found the approach to be novel and the results encouraging, they nevertheless suggested that I execute this research using a more sophisticated methodology and network experimentation. I have been working on this research since and intend to pursue it further.

This paper explores a new framework for lossy image encryption and decryption using a simple shallow encoder neural network E for encryption, and a complex deep decoder neural network D for decryption. E is kept simple so that encoding can be done on low power and portable devices and can in principle be any nonlinear function which outputs an encoded vector. D is trained to decode the encodings using the dataset of image - encoded vector pairs obtained from E and happens independently of E. As the encodings come from E which while being a simple neural network, still has thousands of random parameters and therefore the encodings would be practically impossible to crack without D. This approach differs from autoencoders as D is trained completely independently of E, although the structure may seem similar. Therefore, this paper also explores empirically if a deep neural network can learn to reconstruct the original data in any useful form given the output of a neural network or any other nonlinear function, which can have very useful applications in Cryptanalysis. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the decoded images from D along with some limitations.

• _Led an applied research project to empirically prove that robots (as edge devices) can learn to perform a task (such as walking) using actuators, with both the data being generated, and network trained in real-time using on board sensors and on board computation.

With the advancements in high volume, low precision Computational technology and applied research on cognitive Artificially Intelligent heuristic systems, Machine Learning solutions through Neural Networks with real-time learning has seen an immense interest in the research community as well the Industry. This Paper involves Research, Development and Analysis of a Neural Network implemented on a robot with an arm through which evolves to learn to walk in a straight line or as required. The Neural Network learns using the Algorithms of Gradient Descent and Backpropagation. Both the implementation and training of the Neural Network is done locally on the robot on a Raspberry Pi 3 so that its learning process is completely independent. The Neural Network is first tested on a Simulator developed on MATLAB and then implemented on Raspberry Pi 3. Data at each generation of the evolving network is stored, and analysis both mathematical and graphical is done on the data. Impact of factors like Learning Rate and Error Tolerance on the learning process and final output is analyzed.

• _A modern take on the J.C.R. Licklider’s famous paper on ‘Man-C omputer Symbiosis’ (1960), with a focus on the extent to which current machine learning systems illustrate that symbiosis, and the efficacy of deep learning as a way to achieve true man-computer symbiotic relationships.

Man-Computer Symbiosis (MCS) was originally envisioned by the famous computer pioneer J.C.R. Licklider in 1960, as a logical evolution of the then inchoate relationship between computer and humans. In his paper, Licklider provided a set of criteria by which to judge if a Man-Computer System is a symbiotic one, and also provided some predictions about such systems in the near and far future. Since then, innovations in computer networks and the invention of the Internet were major developments towards that end. However, with most systems based on conventional logical algorithms, many aspects of Licklider's MCS remained unfulfilled. This paper explores the extent to which modern machine learning systems in general, and deep learning ones in particular best exemplify MCS systems, and why they are the prime contenders to achieve a true Man-Computer Symbiosis as described by Licklider in his original paper in the future. The case for deep learning is built by illustrating each point of the original criteria as well as the criteria laid by subsequent research into MCS systems, with specific examples and applications provided to strengthen the arguments. The efficacy of deep neural networks in achieving Artificial General Intelligence, which would be the perfect version of an MCS system is also explored.