Optimal Transport

Professor Cédric Villani, “Optimal Transport Theory”, New Frontiers in Mathematics, 2018/1/17.

Making Better Fulfillment Decisions on the Fly in an Online Retail Environment

Jason Acimovic and Stephen C. Graves, Making Better Fulfillment Decisions on the Fly in an Online Retail Environment, Manufacturing & Service Operations Management, Volume 17, Issue 1, Winter 2015. (2017 M&SOM Best Paper Award)

We develop a heuristic that makes fulfillment decisions by minimizing the immediate outbound shipping cost plus an estimate of future expected outbound shipping costs (*). These estimates are derived from the dual values of a transportation linear program (LP). In our experiments on industry data, we capture 36% of the opportunity gap assuming clairvoyance, leading to reductions in outbound shipping costs on the order of 1%. These cost savings are achieved without any deterioration in customer service levels or any increase in holding costs.

東北角偏鄉小校 雲端教育大爆發

商業周刊，東北角偏鄉小校 雲端教育大爆發，2020年4月24日

Chip Placement with Deep Reinforcement Learning

Azalia Mirhoseini, et al., Chip Placement with Deep Reinforcement Learning, arXiv:2004.10746.

In this work, we present a learning-based approach to chip placement, one of the most complex and time-consuming stages of the chip design process. Unlike prior methods, our approach has the ability to learn from past experience and improve over time. In particular, as we train over a greater number of chip blocks, our method becomes better at rapidly generating optimized placements for previously unseen chip blocks. To achieve these results, we pose placement as a Reinforcement Learning (RL) problem and train an agent to place the nodes of a chip netlist onto a chip canvas. To enable our RL policy to generalize to unseen blocks, we ground representation learning in the supervised task of predicting placement quality. By designing a neural architecture that can accurately predict reward across a wide variety of netlists and their placements, we are able to generate rich feature embeddings of the input netlists. We then use this architecture as the encoder of our policy and value networks to enable transfer learning. Our objective is to minimize PPA (power, performance, and area), and we show that, in under 6 hours, our method can generate placements that are superhuman or comparable on modern accelerator netlists, whereas existing baselines require human experts in the loop and take several weeks.

How to motivate yourself to change your behavior

Tali Sharot, How to motivate yourself to change your behavior, TEDxCambridge, 2014/10/28

How to do it in 3 steps: social incentives, immediate reward, and progress monitoring (at 15:22)