Why We Sleep

If you’ve ever wondered why you are on track to lose nearly a third of your life to sleep, or are not entirely happy about your relationship with sleep, then this book is a must read.  I first heard about it from Sam Harris’s interview of the author, Matthew Walker, a professor of neuroscience and psychology at UC Berkeley, and was immediately intrigued by their conversation. At least this time, my curiosity did pay off, as I have learned so much that I did not know before.

Our sleep consists of two kinds, the kind with dreaming, called rapid-eye movement (REM) sleep, and the kind without, called non-REM (NREM) sleep or deep sleep. They serve different functions. Basically, NREM sleep helps clean the brain, consolidate and retain memory, and therefore is critical to learning and retaining knowledge, as well as maintaining cognitive ability.  Persistent lack of NREM sleep is a known risk factor for Alzheimer’s (old timers) disease.  REM sleep, on the other hand, is closely related to emotion and social behaviors.  The newborns of heavy-drinking mothers are more likely to suffer from mental illness—including autism—when they grow up, partly because their REM sleep are disrupted by alcohol.

Sleep is controlled by two processes: circadian cycle (生物钟)and sleep pressure.  The circadian cycle is an internal clock regulated by melatonin (脑白金), whose main function is to tell the brain and body “it is dark and please get ready for bed”.  Next time when you take melatonin to mitigate your suffering from a jet-lag, remember that message, and that message alone, is what you are getting. The sleep pressure is created by another chemical called adenosine, which begins to build up in your brain once you wake up and which can only be reduced by sleep.  Drinking coffee, however, can resist sleep pressure because caffeine helps block the receptor cells in the brain designed to “feel” the pressure.   Two facts about caffeine are especially noteworthy – sorry coffee drinkers, but please read.  First, “caffeine is one of the most common culprits that keep people from falling asleep easily and sleeping soundly thereafter”. Second, if you cannot get through the morning without caffeine, then most likely you have “self-medicated your state of chronic sleep deprivation”.

Walker is eager to tell everyone who would listen that getting sufficient sleep, at least seven hours a day, is of pivotal importance to human health, neurologically and physiologically. Human beings routinely give up sleep in exchange for activities deemed more productive, valuable, or enjoyable.  In some cultures, self-inflicted sleep deprivation is an emblem of work ethic, if not a badge of honor.   Walker repeatedly warns us of the grieve danger of this chivalrous attitude toward sleep. His book documents, sometimes with gruesome details, how sleep loss could inflict devastating, even lethal, effects on the brain and the body, causing or worsening countless disorders and diseases, ranging from anxiety, depression and bipolar disorder to cancer, diabetes, obesity, and infertility.   Remember, 99% of humans cannot function optimally without at least seven-hour-sleep a day; so obviously your odds of being among that 1% (who has a sub-variant of a gene called BHLHE41) is not as good as you might like.

Let me end with a laundry list of Dos and don’ts. First and foremost, neither sleeping pill nor alcohol can help you sleep better.   As sedatives, these substances give you not so much good sleep as induced unconsciousness.   In other words, you may think you have slept, but you would not get any benefits associated with sleeping.  Here are a few things that do help: (i) reduce caffeine and alcohol intake; (ii) avoid exposure to LED light before sleep (including from screens of your phones, tablets, and computers), (iii) have a cool bedroom (around 18 degree Celsius);  remember, to initiate sleep, your core temperature need to drop about 1 degree Celsius, and finally (iv) stick to a regular bedtime and wake-up time as much as possible.

Bi-criteria traffic assignment

The traffic assignment problem (TAP) aims to find the distribution of agents—travelers, goods or data—in a network according to certain rules that govern how the agents make choices and move in the network. This problem lies at the heart of numerous applications, ranging from infrastructure planning to travel demand management. In these applications, it is often important to differentiate the agents according to the governing rules.  The trade-off between two attributes by agents with heterogeneous preferences is ubiquitous in route choice, traffic assignment, congestion games and beyond, and it leads to the bi-criteria traffic assignment (BiTA) problem concerned herein.  In this study, we develop a novel  algorithm to solve a continuous version of the BiTA problem. See Abstract for details.

This is a joint work with my former visiting PhD student and  Postdoc Jun Xie (currently Associate Professor at Southwest Jiaotong University, Chengdu, China) and Qianni Wang, an MS student currently at Northwestern University. The paper is currently under revision at Operations Research; a preprint can be downloaded here.


Abstract: This paper studies the continuous bi-criteria traffic assignment (C-BiTA) problem, which aims to find the distribution of agents with heterogeneous preferences in a network. The agents can be seen as playing a congestion game and their payoff is a linear combination of time and toll accumulated over the selected path. We rediscover a formulation that enables the development of a novel and highly efficient algorithm. The novelty of the algorithm lies in a decomposition scheme and a special potential function. Together, they reduce a complex assignment problem into a series of single boundary adjustment (SBA) operations, which simply shift flows between “adjacent” efficient paths connecting an origin-destination (OD) pair using a Newton method. The SBA algorithm is capable of producing highly detailed path-based solutions that hitherto are not widely available to C-BiTA. Our numerical experiments, which are performed on networks with up to forty thousand links and millions of OD pairs, confirmed the consistent and significant computational advantage of the SBA algorithm over the Frank-Wolfe (FW) algorithm, the widely-held benchmark for C-BiTA. In most cases, SBA offers a speedup of an order of magnitude.  We also uncovered evidence suggesting the discretization-based approach—or the standard multi-class formulation—-is likely to produce far more used paths per OD pair than C-BiTA, a potential computational disadvantage. Equipped with the proposed algorithm, C-BiTA, as well as its variants and extensions, could become a viable tool for researchers and practitioners seeking to apply multi-criteria assignment models on large networks.