AY 2017 GT Topic Assignments
をテンプレートにして作成
[
トップ
] [
新規
|
一覧
|
単語検索
|
最終更新
|
ヘルプ
|
ログイン
]
開始行:
CENTER:SIZE(60){COLOR(gold){AY 2017 GT Topics }}
CENTER:COLOR(green){''Kick-off seminar on Wednesday, June 14, 2017''}
*COLOR(yellow){NASH Project} - http://adaptive.u-aizu.ac.jp/aslint/index.php?NASH [#ab7b22b9]
----
*** COLOR(blue){GT1-S1: FPGA Acceleration and Performance Study of Character Recognition with Feed-Forward Neural Network&br [順伝搬型ニューラルネットワークを用いた文字認識システムの性能調査とFPGAによる高速化]} [#hf82f035]
-[[''s1220042 Masaki Yamada GT1-S1''>http://adaptive.u-aizu.ac.jp/aslint/index.php?Masaki%20Yamada]]
''Motivation'': Feed Forward Neural Networks (FFNN) when designed to work with floating point (FP) precision performs a large number of elementary products
and sums. For each neuron of FFNN within the hidden layers, a non-linear function computation is required to determine the activation value
of the neuron. Without efficient, dedicated FP hardware, such computations can create difficulties for the whole system performance of the system, hence making the design difficult to be used in critical applications like real-time systems.
''Goal:''The goal of this research is to implement a Feed Forward Neural Networks (FFNN) with floating point on FPGA. A real application, such as character recognition, should be demonstrated.
----
***COLOR(blue){GT1-S2: Study of Character Recognition with Feed-Forward Neural Network} [#hc6edd29]
-[[''s1210201 Toyama GT1-S2''>http://adaptive.u-aizu.ac.jp/aslint/index.php?cmd=read&page=Kota%20Toyama&refer=Members-Internal]]
''Motivation'': Feed Forward Neural Networks (FFNN) when designed to work with floating point (FP) precision performs a large number of elementary products
and sums. For each neuron of FFNN within the hidden layers, a non-linear function computation is required to determine the activation value
of the neuron. Without efficient, dedicated FP hardware, such computations can create difficulties for the whole system performance of the system, hence making the design difficult to be used in critical applications like real-time systems.
''Goal:''The goal of this research is to implement a Feed Forward Neural Networks (FFNN) with floating point on FPGA. A real application, such as character recognition, should be demonstrated. The FFNN should be trained in Matlab environment and the Nios II/f (co cache) should be used for Altera FPGA prototyping. The Nios II ISA should be extended to have a Floating Point ALU.
----
***COLOR(blue){GT2: Design of a Leaky, Integrate and Fire (LIF) Neuron Core for NASH System [NASH システムのための漏れ積分発火ニューロンモデルコアの設計]} [#z106456d]
-[[B4 Kanta Suzuki>Kanta Suzuki]](s1220215)
''Motivation:'' Spiking neural networks (SNN) are a set of neurons that communicate through spikes and compute through the timing of the spike. These spiking neurons have become popular since they mimic the spiking nature of biological neurons and can reproduce those neuron spiking patterns.
The Hodgkin-Huxley model is one of the most detailed and best known models of spiking neurons.
It describes the subcellular level behaviors, the membrane current generation and propagation of neural spikes.
And because of these high level of details, the Hodgkin-Huxley model is too complex to be used for a large scale simulation or hardware implementation.
For large scale hardware implementations, simplified models such as integrate-andfire model are preferred.
This is due to limited hardware resources for any design and same is the case for this thesis.
Such simpler models can emulate the spiking nature of neurons, most of its behaviors and also keep the cost of computation at a comparatively low level.
''Goal:'' In this thesis, the neuron model used is the leaky, integrate and fire model [[[Ref.5>http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6945192]]] which has a lesser level of complexity than Hodgkin-Huxley and Izhikevich, shall be developed.
//-[[2011.Neuromorphic Silicon Neuron Circuits>https://drive.google.com/file/d/0B2HMlO4p7SuweW9QbnJqaFZkaU0/view?usp=sharing]]
//-[[2009.CMOS Integrate and Fire Neuron for Temporal Logarithmic Encoding>https://drive.google.com/file/d/0B2HMlO4p7SuwR2piZ3lqWVZiems/view?usp=sharing]]
----
***COLOR(blue){GT3-S1: Performance Study of %%SNN%%/ANN in Software [%%SNN%%/ANN のソフトウェア実装における性能調査]} [#lecc885a]
-[[''s1220236 Takakuwa GT3-S1''>http://adaptive.u-aizu.ac.jp/aslint/index.php?Kosuke%20Takakuwa]]
''Motivation'': Neural networks are used in many applications. To study their performance, hardware design is ideal. However, it is time consuming and difficult.
''Goal'': The goal of this GT is to use a software simulator and study the performance of NN/SNN system for image recognition.
References:
-[[SpikeNET>https://sccn.ucsd.edu/~arno/spikenet/]]
-[[GENESIS>https://en.wikipedia.org/wiki/GENESIS_(software)]]
-[[NEURON>https://en.wikipedia.org/wiki/Neuron_(software)]]
-[[NEST>https://en.wikipedia.org/wiki/NEST_(software)]]
-[[Nengo>https://en.wikipedia.org/wiki/Neural_Engineering_Object]]
----
***COLOR(blue){GT3-S2: Performance Study of SNN/%%ANN%% in Software [SNN/%%ANN%% のソフトウェア実装における性能調査]} [#lecc885a]
-[[''s1210087 Inoue GT3-S2''>http://adaptive.u-aizu.ac.jp/aslint/index.php?cmd=read&page=Yuga%20Inoue&refer=Members-Internal]]
''Motivation'': Neural networks are used in many applications. To study their performance, hardware design is ideal. However, it is time consuming and difficult.
''Goal'': The goal of this GT is to use a software simulator and study the performance of NN/SNN system for image recognition.
References:
-[[SpikeNET>https://sccn.ucsd.edu/~arno/spikenet/]]
-[[GENESIS>https://en.wikipedia.org/wiki/GENESIS_(software)]]
-[[NEURON>https://en.wikipedia.org/wiki/Neuron_(software)]]
-[[NEST>https://en.wikipedia.org/wiki/NEST_(software)]]
-[[Nengo>https://en.wikipedia.org/wiki/Neural_Engineering_Object]]
*COLOR(yellow){Smart Drone Project} [#pe85bb10]
** Algorithm development [#l01a38b3]
***COLOR(blue){GT4: Vision-based attitude estimation algorithm for embedded system (組み込みシステムのための視覚に基づく姿勢推定アルゴリズムの実装と評価)} [#ha731c70]
''Motivation'': Typical Unmanned aerial vehicle estimates their attitudes using gyro sensor, magnetometer, and accelerometer, and GPS. GPS is used to obtain the absolute position, and others are used to obtain relative movement. However, the frequency of GPS signal is very low to control UAV’s attitude. Image based attitude estimation has a potential to solve the problem since the frame rate of a typical camera up to 120 Hz. This research investigates real-time image-based algorithms for attitude estimation for UAVs using FPGA.
''Goal'': Prototyping of Vision based attitude estimation algorithm by software, evaluation of the complexity and accuracy, and implementation on FPGA
''Resources'':
- https://www.youtube.com/watch?v=pO5EfMkZWng
***COLOR(blue){ GT5: Development of video classification algorithm using TSCDP (TSCDPを用いた動画分類アルゴリズムの実装と評価)} [#x1b130b8]
''Motivation'': Recognizing video scene is important tasks for understanding the situation in robotics, search engines and so on. TSCDP is one of the simple algorithms to extract pre-defined trajectories from movies, and the results containing the sequence of detected trajectories can be used to explain a scene. This research investigates learning algorithms to classify human motions using typical datasets from scene descriptions using TSCDP.
''Goal'': Improvement of recognition accuracy with TSCDP and machine learning, acceleration of TSCDP on GPGPU
''Resources'':
- http://www.u-aizu.ac.jp/research/faculty/detail?cd=1
***COLOR(blue){ GT6: Performance evaluation of deep reinforcement learning algorithms (深層強化学習アルゴリズムの実装と評価)} [#re775be3]
''Motivation'': To establish learning algorithms controlling agents directly from images or movies is one of the challenges in computer science. Deep reinforcement learning is a new approach in the area, which is consisted from two current techniques composed by deep learning and reinforcement learning. An important application of deep reinforcement learning is to learn “How to play game” from images. One of drawback is processing time of the learning; it takes over ten days to learn proper rules from a game even the simple one. In this research, you evaluate the performance of deep reinforcement learning running on general PCs, GPUs, and embedded processors.
''Goal'': Performance evaluation of reinforcement learning algorithms for drone simulators
''Resources'':
- https://arxiv.org/abs/1701.07274
- https://www.slideshare.net/takahirokubo7792/python-openai-gym
** Practical Implementation [#ge270bf0]
***COLOR(blue){ GT7: Implementation of SLAM system for embedded systems (組み込みシステムの為のSLAMシステムの実装と評価) } [#d9e0fd9e]
-
''Motivation'': SLAM is a framework to do mapping and current position estimation simultaneously from sensor data. This research implements a SLAM system using compact and inexpensive sensors for low-cost autonomous vehicles.
''Goals'':
GT7-1: Implementation and evaluation of SLAM algorithm on FPGA SoC.
GT7-2: Implementation and evaluation of SLAM algorithm on Jetson DK2.
''Resources'':
- https://www.youtube.com/watch?v=IMSozUpFFkU
- http://myenigma.hatenablog.com/entry/20150207/1423298357
***COLOR(blue){ GT8: Hardware implementation of sensor data acquisition and motor control system (センサーからのデータ取得とモーターコントロールシステムのハードウェアによる実装と評価) } [#ycbd24bd]
''Motivation'': Current autonomous vehicles obtain much information from sensors, and output pulse width modulated (PWM) signals to motors with proper processing in micro-controllers. However complicated sensor systems require special programming techniques to deal data acquisition from sensors and output signals to motors due to satisfy the real-time requirement. This research develops hardware modules for sensor data acquisition and motor control to solve the difficulty.
''Goal'': Implementation of motor control, sensor data acquisition, and serial interface on FPGA
''Resources'':
- https://www.youtube.com/watch?v=DTOK6CgXRXg
- https://www.youtube.com/watch?v=X-1YMl6aO1g
//***GT: Performance evaluation of Python-based neural network simulator for embedded systems (組み込みシステムの為のPythonによるニューラルネットワークシミュレータの実装と評価) [#y23f9d79]
//''Motivation'': In recent research, biologically plausible neurons and learning mechanisms contribute to a new approach in cognitions and decision making in short time. This research evaluates TensorFlow //and Nengo that is a simulator for large-scale neural systems with general purpose processors, typical GPUs, embedded processors, and a GPU for embedded systems.
//''Goal'': Implementation and evaluation of neural network simulators on Raspberry Pi3 and Jetson TK2
//''Resources'':
//- https://www.youtube.com/watch?v=bXCmD1xL_zg
///- https://www.youtube.com/watch?v=2FmcHiLCwTU
//- http://www.4gamer.net/games/049/G004964/20170307048/
*[[Other GT Topics>http://adaptive.u-aizu.ac.jp/aslint/index.php?Available%20GT/MS%20Theses]] [#nb90e6cb]
----
-s1220042 Yamada GT1
-s1210201 Toyama GT1
-s1220215 Suzuki GT2
-s1220236 Takakuwa GT3
-s1210087 Inoue GT3
-s1220044 Arakawa GT4
-s1220205 Yamaguchi GT5
-s1220017 Fukuchi GT6
-s1220167 Igarashi GT7
-s1230053 Shimmyo GT7
-s1190197 Tanaka GT8
終了行:
CENTER:SIZE(60){COLOR(gold){AY 2017 GT Topics }}
CENTER:COLOR(green){''Kick-off seminar on Wednesday, June 14, 2017''}
*COLOR(yellow){NASH Project} - http://adaptive.u-aizu.ac.jp/aslint/index.php?NASH [#ab7b22b9]
----
*** COLOR(blue){GT1-S1: FPGA Acceleration and Performance Study of Character Recognition with Feed-Forward Neural Network&br [順伝搬型ニューラルネットワークを用いた文字認識システムの性能調査とFPGAによる高速化]} [#hf82f035]
-[[''s1220042 Masaki Yamada GT1-S1''>http://adaptive.u-aizu.ac.jp/aslint/index.php?Masaki%20Yamada]]
''Motivation'': Feed Forward Neural Networks (FFNN) when designed to work with floating point (FP) precision performs a large number of elementary products
and sums. For each neuron of FFNN within the hidden layers, a non-linear function computation is required to determine the activation value
of the neuron. Without efficient, dedicated FP hardware, such computations can create difficulties for the whole system performance of the system, hence making the design difficult to be used in critical applications like real-time systems.
''Goal:''The goal of this research is to implement a Feed Forward Neural Networks (FFNN) with floating point on FPGA. A real application, such as character recognition, should be demonstrated.
----
***COLOR(blue){GT1-S2: Study of Character Recognition with Feed-Forward Neural Network} [#hc6edd29]
-[[''s1210201 Toyama GT1-S2''>http://adaptive.u-aizu.ac.jp/aslint/index.php?cmd=read&page=Kota%20Toyama&refer=Members-Internal]]
''Motivation'': Feed Forward Neural Networks (FFNN) when designed to work with floating point (FP) precision performs a large number of elementary products
and sums. For each neuron of FFNN within the hidden layers, a non-linear function computation is required to determine the activation value
of the neuron. Without efficient, dedicated FP hardware, such computations can create difficulties for the whole system performance of the system, hence making the design difficult to be used in critical applications like real-time systems.
''Goal:''The goal of this research is to implement a Feed Forward Neural Networks (FFNN) with floating point on FPGA. A real application, such as character recognition, should be demonstrated. The FFNN should be trained in Matlab environment and the Nios II/f (co cache) should be used for Altera FPGA prototyping. The Nios II ISA should be extended to have a Floating Point ALU.
----
***COLOR(blue){GT2: Design of a Leaky, Integrate and Fire (LIF) Neuron Core for NASH System [NASH システムのための漏れ積分発火ニューロンモデルコアの設計]} [#z106456d]
-[[B4 Kanta Suzuki>Kanta Suzuki]](s1220215)
''Motivation:'' Spiking neural networks (SNN) are a set of neurons that communicate through spikes and compute through the timing of the spike. These spiking neurons have become popular since they mimic the spiking nature of biological neurons and can reproduce those neuron spiking patterns.
The Hodgkin-Huxley model is one of the most detailed and best known models of spiking neurons.
It describes the subcellular level behaviors, the membrane current generation and propagation of neural spikes.
And because of these high level of details, the Hodgkin-Huxley model is too complex to be used for a large scale simulation or hardware implementation.
For large scale hardware implementations, simplified models such as integrate-andfire model are preferred.
This is due to limited hardware resources for any design and same is the case for this thesis.
Such simpler models can emulate the spiking nature of neurons, most of its behaviors and also keep the cost of computation at a comparatively low level.
''Goal:'' In this thesis, the neuron model used is the leaky, integrate and fire model [[[Ref.5>http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6945192]]] which has a lesser level of complexity than Hodgkin-Huxley and Izhikevich, shall be developed.
//-[[2011.Neuromorphic Silicon Neuron Circuits>https://drive.google.com/file/d/0B2HMlO4p7SuweW9QbnJqaFZkaU0/view?usp=sharing]]
//-[[2009.CMOS Integrate and Fire Neuron for Temporal Logarithmic Encoding>https://drive.google.com/file/d/0B2HMlO4p7SuwR2piZ3lqWVZiems/view?usp=sharing]]
----
***COLOR(blue){GT3-S1: Performance Study of %%SNN%%/ANN in Software [%%SNN%%/ANN のソフトウェア実装における性能調査]} [#lecc885a]
-[[''s1220236 Takakuwa GT3-S1''>http://adaptive.u-aizu.ac.jp/aslint/index.php?Kosuke%20Takakuwa]]
''Motivation'': Neural networks are used in many applications. To study their performance, hardware design is ideal. However, it is time consuming and difficult.
''Goal'': The goal of this GT is to use a software simulator and study the performance of NN/SNN system for image recognition.
References:
-[[SpikeNET>https://sccn.ucsd.edu/~arno/spikenet/]]
-[[GENESIS>https://en.wikipedia.org/wiki/GENESIS_(software)]]
-[[NEURON>https://en.wikipedia.org/wiki/Neuron_(software)]]
-[[NEST>https://en.wikipedia.org/wiki/NEST_(software)]]
-[[Nengo>https://en.wikipedia.org/wiki/Neural_Engineering_Object]]
----
***COLOR(blue){GT3-S2: Performance Study of SNN/%%ANN%% in Software [SNN/%%ANN%% のソフトウェア実装における性能調査]} [#lecc885a]
-[[''s1210087 Inoue GT3-S2''>http://adaptive.u-aizu.ac.jp/aslint/index.php?cmd=read&page=Yuga%20Inoue&refer=Members-Internal]]
''Motivation'': Neural networks are used in many applications. To study their performance, hardware design is ideal. However, it is time consuming and difficult.
''Goal'': The goal of this GT is to use a software simulator and study the performance of NN/SNN system for image recognition.
References:
-[[SpikeNET>https://sccn.ucsd.edu/~arno/spikenet/]]
-[[GENESIS>https://en.wikipedia.org/wiki/GENESIS_(software)]]
-[[NEURON>https://en.wikipedia.org/wiki/Neuron_(software)]]
-[[NEST>https://en.wikipedia.org/wiki/NEST_(software)]]
-[[Nengo>https://en.wikipedia.org/wiki/Neural_Engineering_Object]]
*COLOR(yellow){Smart Drone Project} [#pe85bb10]
** Algorithm development [#l01a38b3]
***COLOR(blue){GT4: Vision-based attitude estimation algorithm for embedded system (組み込みシステムのための視覚に基づく姿勢推定アルゴリズムの実装と評価)} [#ha731c70]
''Motivation'': Typical Unmanned aerial vehicle estimates their attitudes using gyro sensor, magnetometer, and accelerometer, and GPS. GPS is used to obtain the absolute position, and others are used to obtain relative movement. However, the frequency of GPS signal is very low to control UAV’s attitude. Image based attitude estimation has a potential to solve the problem since the frame rate of a typical camera up to 120 Hz. This research investigates real-time image-based algorithms for attitude estimation for UAVs using FPGA.
''Goal'': Prototyping of Vision based attitude estimation algorithm by software, evaluation of the complexity and accuracy, and implementation on FPGA
''Resources'':
- https://www.youtube.com/watch?v=pO5EfMkZWng
***COLOR(blue){ GT5: Development of video classification algorithm using TSCDP (TSCDPを用いた動画分類アルゴリズムの実装と評価)} [#x1b130b8]
''Motivation'': Recognizing video scene is important tasks for understanding the situation in robotics, search engines and so on. TSCDP is one of the simple algorithms to extract pre-defined trajectories from movies, and the results containing the sequence of detected trajectories can be used to explain a scene. This research investigates learning algorithms to classify human motions using typical datasets from scene descriptions using TSCDP.
''Goal'': Improvement of recognition accuracy with TSCDP and machine learning, acceleration of TSCDP on GPGPU
''Resources'':
- http://www.u-aizu.ac.jp/research/faculty/detail?cd=1
***COLOR(blue){ GT6: Performance evaluation of deep reinforcement learning algorithms (深層強化学習アルゴリズムの実装と評価)} [#re775be3]
''Motivation'': To establish learning algorithms controlling agents directly from images or movies is one of the challenges in computer science. Deep reinforcement learning is a new approach in the area, which is consisted from two current techniques composed by deep learning and reinforcement learning. An important application of deep reinforcement learning is to learn “How to play game” from images. One of drawback is processing time of the learning; it takes over ten days to learn proper rules from a game even the simple one. In this research, you evaluate the performance of deep reinforcement learning running on general PCs, GPUs, and embedded processors.
''Goal'': Performance evaluation of reinforcement learning algorithms for drone simulators
''Resources'':
- https://arxiv.org/abs/1701.07274
- https://www.slideshare.net/takahirokubo7792/python-openai-gym
** Practical Implementation [#ge270bf0]
***COLOR(blue){ GT7: Implementation of SLAM system for embedded systems (組み込みシステムの為のSLAMシステムの実装と評価) } [#d9e0fd9e]
-
''Motivation'': SLAM is a framework to do mapping and current position estimation simultaneously from sensor data. This research implements a SLAM system using compact and inexpensive sensors for low-cost autonomous vehicles.
''Goals'':
GT7-1: Implementation and evaluation of SLAM algorithm on FPGA SoC.
GT7-2: Implementation and evaluation of SLAM algorithm on Jetson DK2.
''Resources'':
- https://www.youtube.com/watch?v=IMSozUpFFkU
- http://myenigma.hatenablog.com/entry/20150207/1423298357
***COLOR(blue){ GT8: Hardware implementation of sensor data acquisition and motor control system (センサーからのデータ取得とモーターコントロールシステムのハードウェアによる実装と評価) } [#ycbd24bd]
''Motivation'': Current autonomous vehicles obtain much information from sensors, and output pulse width modulated (PWM) signals to motors with proper processing in micro-controllers. However complicated sensor systems require special programming techniques to deal data acquisition from sensors and output signals to motors due to satisfy the real-time requirement. This research develops hardware modules for sensor data acquisition and motor control to solve the difficulty.
''Goal'': Implementation of motor control, sensor data acquisition, and serial interface on FPGA
''Resources'':
- https://www.youtube.com/watch?v=DTOK6CgXRXg
- https://www.youtube.com/watch?v=X-1YMl6aO1g
//***GT: Performance evaluation of Python-based neural network simulator for embedded systems (組み込みシステムの為のPythonによるニューラルネットワークシミュレータの実装と評価) [#y23f9d79]
//''Motivation'': In recent research, biologically plausible neurons and learning mechanisms contribute to a new approach in cognitions and decision making in short time. This research evaluates TensorFlow //and Nengo that is a simulator for large-scale neural systems with general purpose processors, typical GPUs, embedded processors, and a GPU for embedded systems.
//''Goal'': Implementation and evaluation of neural network simulators on Raspberry Pi3 and Jetson TK2
//''Resources'':
//- https://www.youtube.com/watch?v=bXCmD1xL_zg
///- https://www.youtube.com/watch?v=2FmcHiLCwTU
//- http://www.4gamer.net/games/049/G004964/20170307048/
*[[Other GT Topics>http://adaptive.u-aizu.ac.jp/aslint/index.php?Available%20GT/MS%20Theses]] [#nb90e6cb]
----
-s1220042 Yamada GT1
-s1210201 Toyama GT1
-s1220215 Suzuki GT2
-s1220236 Takakuwa GT3
-s1210087 Inoue GT3
-s1220044 Arakawa GT4
-s1220205 Yamaguchi GT5
-s1220017 Fukuchi GT6
-s1220167 Igarashi GT7
-s1230053 Shimmyo GT7
-s1190197 Tanaka GT8
ページ名: