Differences

This shows you the differences between two versions of the page.

--- mobility_data:top [2022/03/10 08:53] – [Meeting Note] zhiyuanpeng
+++ mobility_data:top [2022/10/24 09:24] (current) – [Meeting Note] zhiyuanpeng
@@ Line 1: / Line 1: @@
-====== 2021 Project Page ======
+====== 2021-2022 Project Page ======
 ===== Mobility Data Project =====
 ==== Meeting Note ====
+/10/24
+  * Attendee: Zhiyuan, Xiangyu, Yuanbo, Yang
+  * Meeting Summary
+  Xiangyu
+  * Xiangyu shared a instance normalization method for time-series forecasting against distribution shift which published in ICLR 2022.
+    * Using this method the MSE of MLP for gaode's dataset(350days training data,50days test data) improved from 4.65 to 3.26( 30%+).
+    * Next step I will implement this method in our meta-learning framewrok to see the improvements and compare the effiectiveness of our method and the normalization.[Presentation-slides:https://cloud.tsinghua.edu.cn/f/75320708a08f404e8a0b/]
+* Zhiyuan
+  * Summary:
+    * this week, I conducted a series of experiments to compare our Soft-restricted MF Multi-task learning model performance with single loss trained ones.
+    * This weeks experiments reveal that the multitask loss only contributes limitedly to the improvement on the both two tasks.
+    * Moreover, the LSTM based backbone tends to predict more smoother compared to the more fluctuated data in reality.
+    * An important observation: MSE mainly comes from several regions that has great volatility.
+  * Future Plan:
+    * Maybe next week we can try to use multi-source input data or time sequence analysis method to deal with it.
+/6/30
+  * Attendee: Zhiyuan, Xiangyu, Yang
+  * Meeting Summary:
+     * Transcribe the matrixed-version formulation into its Lagrange version
+     * Use gradient descend and integer projection to iteratively update the optimization problem
+     * Interpretation and analysis the loss result of the proposed method
+     * Regard the greedy algorithm as the supremum while the dynamic programming and real-relaxed optimization as the infimum and compute their approximation rate to measure the outcome with the optimal result
+     * Integer rounding the result in real-domain by the hyper parameter threshold.
+     * Make sure the application of the Pathlet is use a top-k dictionary to recomposite new trajectory and find how much information can be kept
+  * TO-DO:
+     * Find the optimal threshold to integer round the real-domain solution, meaning we use matrix formulation to update result while guaranteeing the integer constraints
+     * Find the application for the Pathlet in real-world project such as new coding for the spatial data
+/06/08
+  * Attendee: Zhiyuan, Yuanbo, Yang
+  * Meeting Summary:
+  * Zhiyuan Peng:
+     * Briefly introduce the general implementation of the dynamic programming with specific data structure.
+     * Exemplify one trajectory results under different λ.
+  * Yuanbo Tang
+     * Introduce the proposed algorithm in state machine term and analogues to the backpack problem in greedy thoughts (but need to connected to a specific math problem in order for its correctness)
+     * Analyze and compare the result of proposed method with DP (including dictionary size and representation cost under different λ)
+     * Use toy model to interpret the logic fallacy and disability to the optimal results
+     * Based on finite map and sufficient trajectories, Transcribe the problem into matrix version
+  * TO-DO:
+     * Survey on the linear restricted minmax convex optimization and whether the max option can be replaced?
+     * Think the optimization in matrix version, proposed optimization algorithm version. Could it be connected with Markov decision process
+     * If our model brings in more features, we can use NN.
+     * Connect the proposed algorithm with some math problem like multi-armed bandit
+/05/11
+  * Attendee: Zhiyuan, Xiangyu, Yang
+  * Meeting Summary:
+     * Use package ‘pyechart’ in python to visualize the cumulative record amount in different h3 hexagon with data from XIAN
+     * Implement LSTM on the processed data to predict the future record in each 5min time piece till 5 times
+     * Prepare the PPT for opening report
+  * To-Do
+     * Find the historical data construction used in literature，survey on how to jointly use long/short term data. We can use fusion or residue methods
+     * Find the metrics to measure the prediction other than MSE, such as MAPE
+     * For solution of the data sparsity and erratic, we can make a hybrid model.
+        * First, we set a threshold to filter out those districts of historical average less than the threshold, and adopt a traditional statistical model
+        * Second, we can use our multi-task DNN to predict other places.
+        * At last, we can hybrid these two situations together
+     * Find out the challenge and remedial algorithms
 /03/09
   * Attendee: Zhiyuan, Xiangyu, Yang
@@ Line 7: / Line 72: @@
     * Zhiyuan: Share the LSTM+GCN overfitting problems, Yang gives some advice:
       * 1. Do more data mining in dataset such as visualize the prediction results and see the correlation about the closeness points in the graph.
     * Xiangyu: Show the results about META-MLP and MLP in UberNYC. Yang gives advice:
       * 1. Divide the task into weekdays and weekends looks like a cross-domain problem. In cross-domain problems, the pre-training method is often better than meta-learning, share the article about the cross-domain.