About
We develop novel and lean technology for general time series forecasting, and pervasive mobile, cloud computing. We care about the following problems.
How to forecast
We consider that any two objects may boast of predictive power for each other. The problem is to compute the conditional distribution of ’s change in the future given ’s change at present. We build TS Terminal.
Koenker, R. (2005). Quantile Regression (Econometric Society Monographs). Cambridge: Cambridge University Press. doi:10.1017/CBO9780511754098
MaxSave
We are raising for maxsave.net, individual-facing platform for the long-term saving and investment need in the UK. It will be content-led on the front page, targeting people looking for finanical knowledge, offering insights about market driving forces for up to one year ahead.
How to compute: a unified look
We see two phases: process current information into memory, then based on the memory, generates by self what to attend to in the next step, and receive new information (from memory or outside) in the next step. The three architectures are fundamentally very similar:
- Turing machine
- attention-based mechanism
- Long-short term memory
Lean hardware and software
We favour simple hardware and software. Something a high schooler can play with.