| 研究課題/領域番号 |
15K15981
|
|---|
| 研究機関 | 奈良先端科学技術大学院大学 |
|---|
研究代表者 |
高 俊涛 奈良先端科学技術大学院大学, 情報科学研究科, 助教 (30732961)
|
|---|
| 研究期間 (年度) |
2015-04-01 – 2018-03-31
|
| キーワード | secure routing / byzantine attack / adaptive trasmission / power constraint / traffic scheduling |
|---|
| 研究実績の概要 |
This year we proposed a monitoring scheme to secure packets route against Byzantine attacks in mobile payment systems; a adaptive packet transmission algorithm which only transmits packets when wireless channel condition is good and does not transmit packets when wireless channel condition is bad, thus saving battery power; and finally a back-pressure based traffic scheduling algorithm that enables bank vehicles/agents to deliver electric money/goods effectively to mobile payment system users.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Reusing our previous methods/knowledge and simulation codes are the two major reasons that our research proceeded smoothly. Specifically, for secure communication research, we reused the security technique, monitoring mechanism, proposed in mobile payment system to prevent byzantine attack; for saving battery power of smart phones, we formulated the problem as an optimization problem with constraints and reused our knowledge of lyapunov optimization method to solve it; for delivering electric money/goods effectively from a bank to disaster areas, we surveyed the latest research advances in traffic scheduling that can reduce vehicle delay in road networks and found that back-pressure based algorithm is promising. We modified previous simulation codes to verify our new proposed schemes.
|
| 今後の研究の推進方策 |
We plan to apply deep learning algorithms to further improve our mobile payment system. To improve system security, we are going to design deep learning algorithms that can learn various attack patterns from training data with labels and then detect attacks and warn users of these attacks. Since labeling training data costs a lot of human power and time, the second step is to design deep learning algorithms that can learn attack patterns from training data without labels, one promising method to achieve this is to combine deep learning and reinforcement learning. Another future work is to apply deep learning for priority traffic signal control and vehicle routing such that bank agents of mobile payment system can reach disaster areas quickly.
|
| 次年度使用額が生じた理由 |
1) To develop deep learning based algorithms, we need to train deep neural networks with data of high dimensions, which requires high performance computers especially with GPU computing capabilities. 2) We plan to attend two domestic conferences and two international conferences, which involve conference registration accommodation, food, transportation fees. 3) We will also submit our work to an international journal, summarizing our proposed mobile payment system, which requires publication fee.
|
| 次年度使用額の使用計画 |
1) Buying GPGPU workstation GU-1000: 300,000 Yen 2) Two domestic conferences: 200,000 Yen 3) Two international conferences: 500,000 Yen 4) one international journal: 100,000 Yen
|
