Deadline to register is Jun 30, 2019.
Teams can still edit your proposals during judging period.

Category: Sort By:
Regional Final
📁Machine Learning
👤Dhruvi Bhadiadra (Portland State University)
📅Jul 02, 2019
As the inventions in the field of Deep Neural Network and their corresponding application explodes day-by-day, the size of the deep neural network also increases rapidly for complicated application. There are concepts like averaging or augmenting three deep neural networks to form the desired classification application. Even though the GPU can be used to pace up the training and inference process of the model by using the multiple thread processing available, after training the memory used to store the model is in terms of hundreds of MegaBytes. So, in the past few years, a separate path of research started on how to reduce the memory used in the storage of parameters. We chose to follow one of the papers[1] and implement the design in hardware for the Proof of Concept. The reason we chose the implementation of hardware is to research the scope of deep learning algorithm running on hardware with low power consumption and high parallelism. The resource availability in the hardware is one of the main struggles we face during the design phase. So, we compress the deep learning parameters using the method provided in the paper and in turn implement the compressed model in FPGA.
details »
2
votes

📁Machine Learning
👤Priyanka K (San Jose State Univesity )
📅Jun 26, 2019
Deep learning is a trending technology, which is beyond the shallow machine learning. Aim of our project is to construct the image with the details given through voice, that is voice to image translation with the help of deep learning concepts and algorithms. The project aims to use the high-performance computing algorithm and interfacing capabilities of open VINO starter kit FPGA to implement efficient neural network model.

Voice to image conversion finds applications in health care, education tools, social media. The image construction from the speech of patient serves as a diagnostic aid. The ability to create visual images based on the context of speech helps in better delivery of educational content. This is also useful to visualize music and in construction of facial emotions from the context of voice
details »
1
votes

📁Machine Learning
👤Euclides Chuma (Unicamp)
📅Mar 20, 2019
The project consists of developing a platform to identify chemical materials using microwave sensors. This project will use deep learning resources to recognize patterns of dielectric properties of the materials under test to identify which are chemical compounds of these materials.
details »
45
votes

📁Internet of Things
👤Mohamed El-Hadedy (California Polytechnic State University)
📅Jun 30, 2019
We teamed up with partners from NASA to use the dynamic re-configuration for implementing the open-source flight control software Fprime. So far the software was implemented by NASA/JPL on raspberry Pi and we are looking for exploring different way on using FPGA to make the software available on a wide range of the reconfigurable computing platforms.
details »
4
votes

📁Internet of Things
👤Mohamed El-Hadedy (California State Polytechnic University Pomona)
📅Jun 30, 2019
NIST has just launched a new Lightweight cryptographic standardization competition for the small devices. The team will be implementing one of the NIST competition candidate algorithms (gage and engage) on FPGA targeting fewer resources on the chip with acceptable performance aligned with the requirements of the small device. Both OpenCL and Custom System-Verilog tools will be involved,
details »
41
votes

📁Internet of Things
👤Gustavo Bertoli (Instituto Tecnológico de Aeronáutica (ITA))
📅May 20, 2019
It is a trend to deploy Machine Learning (ML) algorithms on the Edge to overcome latency and availability issues related to static and cloud-based architectures.

For specific application of ML for Security the amount of data and latency requires that these solutions be available on the Edge instead of tied with a cloud infrastructure.
details »
4
votes

📁Machine Learning
👤Jianwei Zhang (Arizona State University)
📅May 14, 2019
According to WHO, more than 36 million people in the world are blind. Vision impairment people are tough to go out alone. Some of them can accompany with guide dogs, but most can only use guide sticks. In some countries and regions, even the blind roads and other facilities for the vision impairment people are not adequately maintained. The audio guide for vision impairment people can solve this problem to some extent. The audio guide can provide a lot of useful information, such as traffic lights status, obstacle information and movement status, collision warning, and so on. It can also offer a lot of help in daily life, such as surrounding environment description, item description, figure description, text reading. In this project, the first object is traffic lights status identification and classification. I hope that more features will be added to this audio guide in the future to provide more aids for vision impairment people.
details »
1
votes

📁High Performance Computing
👤Max Marrone (New York University Tandon School of Engineering)
📅May 18, 2019
What if every processor in a massively parallel system could pass messages to every other processor, arbitrarily? This would loosen how "independent" each task must be for a problem to be parallelizable. For example, in a graph processing task, each core could "be" one graph node, even if the graph is highly irregular, with nodes having an arbitrary number of neighbors.

In the 1980s, the Connection Machine supercomputers explored this idea, and were successfully used for diverse applications, including scientific computing and AI research. I aim to implement a general-purpose accelerator inspired by the Connection Machine architecture.
details »
2
votes

📁High Performance Computing
👤Terry Carpenter (Terry Technologies)
📅May 18, 2019
Portable data collection unit for high frame rate imaging.
details »
1
votes

📁High Performance Computing
👤Cosmin Iorga (NoiseCoupling.com)
📅May 26, 2019
This project will focus on developing a prototype of an affordable wearable medical Magnetic Resonance Imaging (MRI) machine that can become available to any doctor’s office. This prototype will consist of multiple wearable MRI detectors that can be attached to the human body and an artificial intelligence processing unit implemented in the Cyclone V GX FPGA of the OpenVINO starter kit. The artificial intelligence algorithm for data processing will be implemented using the OpenCL HPC development platform. Traditional MRI machines can only measure images of motionless objects as they use multiple large magnetic coils that have to be mounted inside the magnetic scanners at fixed locations so that they don’t interfere with each other. While these traditional MRI machines provide significant information for medical diagnosis, there are medical conditions where the dynamics of the moving joints and human body tissues are essential for a complete diagnosis. The wearable MRI prototype project addresses this issue by providing a low cost solution to MRI imaging of the human body while engaged in moving activities.
details »
1
votes

📁Other: Telecommunication
👤Evgeny Ryzhov (DigitalRadio)
📅Jun 22, 2019
MIMO-OFDM transceiver (wideband transceiver with high efficient bandwidth used)
details »
0
votes

📁Other: Computer Vision
👤Grant Yu (Northwestern University)
📅Jun 30, 2019
We attempt to design a real-time face detection system on an FPGA, based on the Viola-Jones algorithm (eponymously created in 2001). The volume of computation that is generally required by face detection systems is a major challenge in terms of implementation. Power, area, and performance are critical areas that demand consideration and optimization. An FPGA implementation of face detection is able to provide comparatively superior hardware resource efficiency, with respect to other architectures such as CPU/GPU. The Viola-Jones algorithm is ideal for an FPGA implementation due to the nature of computation; only adders, comparators, and shifters are used (no multipliers necessary). Using a Terasic DE10-Nano board, a camera module (such as the Terasic D8M-GPIO), HDMI/VGA monitor, and HDMI/VGA cable, a system can be constructed so as to demonstrate fast, hardware-efficient, low-power face detection. A high level implementation of the Viola-Jones shall be used to train the classifier offline (and subsequently obtain the weights).
details »
0
votes


1 2 3 4