Study on Mobile Data 3D Offloading using Deep Reinforcement Learning
| Project/Area Number |
17H01730
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Information network
|
|---|
| Research Institution | Shizuoka University |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
水野 忠則 愛知工業大学, 情報科学部, 教授 (80252162)
|
|---|
| Project Period (FY) |
2017-04-01 – 2021-03-31
|
| Project Status |
Completed (Fiscal Year 2020)
|
| Budget Amount *help |
¥18,590,000 (Direct Cost: ¥14,300,000、Indirect Cost: ¥4,290,000)
Fiscal Year 2020: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
Fiscal Year 2019: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
Fiscal Year 2018: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
Fiscal Year 2017: ¥4,940,000 (Direct Cost: ¥3,800,000、Indirect Cost: ¥1,140,000)
|
|---|
| Keywords | モバイルネットワーク / データオフローディング / 深層強化学習 / モバイル / オフローディング |
|---|
| Outline of Final Research Achievements |
Regarding the evaluation of mobile data 3D offloading aiming at maximizing space utilization efficiency, under what conditions and how the UE should transmit data to appropriately smooth the eNB load, we evaluated applying deep reinforcement learning to network simulation with various condition settings. In addition to the evaluation of transmission rate control models using DQN, we proceeded with the research focusing on the acquisition of effective communication control methods for 5G network slicing management. Based on the design that allocate network resources to one slice by distributed learning using Ape-X, we confirmed it was possible to optimally allocate resources independently of the number of slices by changing the number of agents.
|
| Academic Significance and Societal Importance of the Research Achievements |
様々な遅延耐性を持つコンテンツの特性を活かしたモバイル3Dオフローディング手法に対し，時間的，空間的局所性を考慮して空間利用効率の最大化を図る制御手法を深層強化学習によって獲得可能なことを示した．実機での適切な評価が規模的に困難かつ，解析モデルやネットワークシミュレーションによる評価では条件設定やモデル構築を現実に近づけるのが困難な情報ネットワーク研究に対し，深層強化学習適用の効果を検証した意義は大きい．シミュレータによって得られた強化学習結果を基に実機制御を行い，その実機での結果を基に深層学習を段階的に行うスパイラル学習法が今後ますます重要になると考える．
|
Report
(5 results)
Research Products
(28 results)
