Meeting notes
=====
## 敦聿
::: warning
**2020/03/16**
- [ ] write 5G midterm report

**2020/03/02**
- [ ] write pseudocode of the greedy algorithm
- [ ] List parameters
- node number (proportion of interfering UE,2 cells with different UE number)
- traffic demand distribution
- Cell range (distance)
- RB number (need EE's data)

**2020/02/26**
- [x] list comparison schemes (send me on Fri)
- [x] list different parameters (send me on Fri)
- [ ] Globalcom due: 4/15 (https://globecom2020.ieee-globecom.org/)
- [ ] Related work
- [ ] problem definition

**2020/01/18**
- [x] 01/31: Resource allocation solver
- [x] 01/31: Find greedy algorithm for RA
- [ ] ~~03/01: Paper writing~~
:::

## 又瑄
:::danger
**2020/03/16**
- [ ] implement the modified greedy algorithm

**2020/02/26**
- [x] simulation over time
- [x] gap of beam (remove adjecent beams of a selected beam)

**2020/01/18**
- [x] simulation: data rate, utilization, impact of imperfect SINR
- [x] number of users
- [x] number of beams
- [x] number of antennas
- [x] beamwidth
- [ ] traffic patterns
- [x] check # of antenna, beam width
- [x] fix total power
- [x] 1/31: WCNC camera ready
- [x] 1/31: Read beam estimation paper
:::
## 陳洋
::: success
**2020/03/23**
- [ ] impact of adaption
- [ ] calculate the average rejection rate (x-axis: update interval)
- [ ] solver with sampling
- [ ] sampling rate vs. execution time
- [ ] future plane: compare (rejection rate + prediction errors)

**2020/03/16**
- [x] impact of adaptation (with adapt vs. without adapt)
- [x] impact of epoch interval size
- [x] comparison with upper bound (solver every small interval)

**2020/03/09**
- [x] model accuracy (training and inference)
- [ ] replace model based on utilization and rejection rate

**2020/03/02**
- [x] utilization of each model
- [x] adaptation algorithm
- [x] model distribution based on flow prefix
- [x] find binary accuracy of 20-fold classifier

**2020/02/26**
- [x] request arrival rate of each type of model
- [x] binary classifier in switches (compare accuracy and rejectionr rate)

**2020/02/20**
- [x] historical flows and futures flows
- [x] experiment: acceptance rate and latency
- [x] implement 2-phase inference
- [x] present teacher-student models

**2020/01/18**
- [x] 1/31: simulation: rejection rate, path length
- [x] impact of switch capacity
- [x] impact of flow numbers
- [x] impact of model assignment (y-axis: delay)
- [x] queueing delay
- [x] Read teacher-student models

**2020/02/06**
- [ ] Binary classifier: confusion matrix
- [ ] paper title: Distilling the Knowledge in a Neural Network
:::

## 霈菱
::: info
**2020/03/23**
- [ ] implement algorithm
- [ ] find an example

**2020/03/16**
- [x] update algorithm
- [NFP Algorithm](https://hackmd.io/ktfdUu1aTcOrntfHN9288Q?view)
- [x] implement algorithm

**2020/03/09**
- [x] update algorithm

**2020/03/02**
- [x] List your simulation configurations here
- F: number of flows (20, 40, 60, 80, 100)
- L: Length of chains ([1,4], [1,6], [1,8], [1,10])
- NN: Number of nodes (10, 20, 30, 40, 50)
- C: node capacity (2, 3, 4, 5)
- [ ] Fig 1~5: F(20-100), L=4, NN=10~50, C=2
- [ ] Fig 6~10: F(20-100), L=6, NN=10~50, C=2
- [ ] Fig 11~15: F(20-100), L=8, NN=10~50, C=2
- [ ] Fig 16~20: F(20-100), L=10, NN=10~50, C=2
- [ ] Fig 21~25: L(4,6,8,10), F=20, NN=10~50, C=2
- [ ] Fig 26~30: L(4,6,8,10), F=40, NN=10~50, C=2
- [ ] Fig 31~35: L(4,6,8,10), F=60, NN=10~50, C=2
- [ ] Fig 36~40: L(4,6,8,10), F=80, NN=10~50, C=2
- [ ] Fig 41~45: L(4,6,8,10), F=100, NN=10~50, C=2
- [ ] Fig 46~70: C(2,3,4,5), F=20~100, L:4, NN=10~50
- [ ] Fig 71~95: C(2,3,4,5), F=20~100, L:6, NN=10~50
- [ ] Fig 96~120: C(2,3,4,5), F=20~100, L:8, NN=10~50
- [ ] Fig 121~145: C(2,3,4,5), F=20~100, L:10, NN=10~50
- [x] Think about the algorithm for NFP

**2020/02/20**
- [x] variables:
- [x] Default settings:
- [x] number of flows [1,5]
- [x] length of chain ([1,4], [1,6], [1,8], [1,10]
- [x] number of nodes
- [x] node capacity (number of VMs per node)
- [x] read experiments in the related work

**2020/01/18**
- [x] 1/31: implement the routing algorithm
- [ ] 2/28: find algorithm for parallel NF
:::

## 立為
::: warning
**2020/03/23**
- [ ] Modify the gesture recognition method of two tags
* Probability of two tags add up
- [ ] Avoid reading a tag after successful pairing

**2020/03/16**
- [x] gesture accuracy vs diff number of tags
- [x] report of evaluation results

**2020/03/09**
- [x] 4 users X N tags(10, 15, 20)
- [x] accuracy of tag pairing(4 mag-tag and non-mag-tag X N object tags(5, 10, 15, 20))
- [x] sample / tag(mag-tag and non-mag-tag)
- [x] DTW of distance(mag-tag and non-mag-tag)

**2020/02/26**
- [ ] Title of presented paper:
- [ ] write the outline of the paper
- [ ] compare mag-tag to non-mag-tag
- [x] accuracy of tag pairing (1 user with K objects / k users with 10 objects)
- [x] impact of distance
- [x] number of recevied responces
- [ ] DTW distance for non-mag-tag and mag-tag
- [x] accuracy of gesture recognition (mag-tag and non-mag-tag)
- [x] experiments in the related work (http://graphics.im.ntu.edu.tw/~robin/docs/uist18_liang.pdf)
- [x] evaluation: varying speed and gesture sizes

**2020/02/06**
- [x] 2/13: try smaller number of windows

**2020/01/18**
- [x] 1/31: experiments: 8 gestures
- [x] 1/31: try input without RSSI
- [x] 1/31: try inference using the moving window and take majority vote
:::
## 暐倫
::: danger
**2020/02/24**
- [ ] truncate sequential data, velocity
- [ ] add uncertainty
- [ ] inference
- [x] finalize RL presentation schedule

**2020/02/18**
- [x] implement map inference
- [x] bayesian neural network (uniform)
- [x] restart study group

**2020/01/18**
- [x] 1/31: Read IRL and give a detailed presentation
- [x] 2/10: Trace IRL example code
- [x] Implement IRL-based anomaly detection
:::

## 千鈞
::: success
**2020/02/24**
- [ ] implement RL service chain assignment
- [x] paper title:
- Efficient Provision of Service Function Chains in Overlay Networks using Reinforcement Learning
- Published in: IEEE Transactions on Cloud Computing
- Author(s): Guanglei Li ; Huachun Zhou ; Bohao Feng ; Yuming Zhang ; Shui Yu

**2020/02/17**
- [x] read RL service chain paper
- [ ] continue study group for RL

**2020/01/18**
- [x] 1/31: trace the RL example code (understand the input format, the model architecture and gradient calculation)
- [x] paper reading
:::

## 謙仁
::: info
**2020/03/04**
- [ ] reimplement 又瑄's work
- [x] RB over time with N subcarrier
- [x] generate traffic pattern
- [x] assume each RB has a fixed data rate (fixed number of bits)
- [ ] Paper reading
- [x] Intelligent Reflecting Surface Enhanced Wireless Network via Joint Active and Passive Beamforming
- [ ] Content-Centric Sparse Multicast Beamforming for Cache-Enabled Cloud RAN

**2020/01/18**
- [x] 1/31: implement multi-cell beam steering and calculate SINR
- [x] paper reading
- [x] read 又瑄's work
:::

## 嘉容
::: warning
**2020/03/04**
- [ ] check BLE frequency
- [ ] duration
- [ ] center frequency
- [ ] use two USRPs to receive simultaneously
- [ ] check sampling rate

**2020/02/24**
- [x] implement the denoising model
- [x] recore BT signals

**2020/01/18**
- [x] 1/31: understand Matlab OFDM example codes
- [x] generate bluetooth signals
- [x] record bluetooth signals
- [x] paper reading
:::

## 杜津
::: danger
**2020/03/18**
- [ ] packet size: normal distribution with mean and variance
- [ ] train binary classfier using DNN (70% training / 30% testing)

**2020/03/04**
- [x] clustering
- [ ] paper reading
- [ ] https://dl.acm.org/doi/abs/10.1145/2942358.2942367
- [ ] https://ieeexplore.ieee.org/abstract/document/7784423/
- [ ] https://dl.acm.org/doi/abs/10.1145/3230543.3230551
- [ ] https://dl.acm.org/doi/abs/10.1145/3230543.3230569

**2020/02/24**
- [x] calculate accuracy in a different way
- [x] implement machine clustering
- [x] normalize feature vectors

**2020/01/18**
- [x] 1/31: monitor the flow/coflow completion time
- [x] 2/10: adjust the flow size based on the bandwidth of Mininet
- [x] paper reading
:::