Adaptive Droplet Routing in Digital Microfluidic Biochips Using Deep Reinforcement Learning
Authors: Tung-Che Liang, Zhanwei Zhong, Yaas Bigdeli, Tsung-Yi Ho, Krishnendu Chakrabarty, Richard Fair
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To evaluate our RL framework, we considered seven DMFBs with the number of electrodes ranging from 25 to 1, 225. For each DMFB, we first trained three models with the same network architecture (as described in Table 1) using DMFB-Env, and the models were trained in the healthy mode to achieve the same performance as that of the baseline (Zhao & Chakrabarty, 2012). After training, we evaluated the performance of the models in the degrading mode of DMFB-Env. We also evaluated the RL framework by executing an epigenetic bioassay on a fabricated biochip. |
| Researcher Affiliation | Academia | 1Department of Electrical and Computer Engineering, Duke University, Durham, NC, USA 2Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan. |
| Pseudocode | No | The paper describes methods and algorithms (e.g., PPO) but does not provide pseudocode or a clearly labeled algorithm block. |
| Open Source Code | Yes | We develop a DMFB simulator in Open AI Gym environment. We open-source the simulator to the RL community for future research1. 1https://github.com/tcliang-tw/dmfb-env. git |
| Open Datasets | No | The paper uses its own simulated environment ('DMFB-Env') to generate random routing tasks for training. It does not use a pre-existing publicly available dataset or provide access information for the generated training data. |
| Dataset Splits | No | The paper describes training and evaluating its model within a simulated environment, but it does not specify explicit training/validation/test dataset splits of a fixed dataset. |
| Hardware Specification | No | The paper describes the hardware for the fabricated DMFB (e.g., Raspberry Pi 3B+, ARMv7 processor, CCD camera, relay ICs) for the physical experiment and deployment, but it does not specify the computational hardware (e.g., GPUs, CPUs, or memory) used to train the deep reinforcement learning models. |
| Software Dependencies | No | The paper mentions using 'Open AI Gym Interface' and the 'proximal policy optimization (PPO) algorithm' but does not specify version numbers for these or any other software libraries or frameworks used for implementation. |
| Experiment Setup | Yes | We tested two significant parameters in PPO to find the best performance of our RL agent for different sizes of DMFBs, the number of concurrent environments, and the number of steps for each update. ... We chose eight concurrent environments as the PPO setting for model training. ... The convolutional neural network configuration. Layer Type depth Activation Stride Padding 1 Convolution 32 Re LU 3 1 2 Convolution 64 Re LU 3 1 3 Convolution 64 Re LU 3 1 4 Fully-Connected 256 Re LU N/A N/A 5 Fully-Connected 4 Re LU N/A N/A |