Feng Ling

PauseRestart hot

Ecosystems, economies and societies are made of many different agents that learn and pursue their own goals. They are quite unlike the sandpiles of classic self-organized criticality. To ask how such systems organise themselves, we built a minimal model called EvoSK. Its agents learn by reinforcement on a rugged spin-glass landscape, and again and again the least fit agent is replaced by a newcomer with a random exploration temperature.

Without any tuning, the population evolves to the edge of ergodicity breaking of its own learning dynamics. There it shows scale-free avalanches of change, and it reaches a collective reward higher than any fixed, hand-tuned temperature can give. These three findings all follow from one condition: the cascade of changes set off by each replacement branches marginally, with a branching ratio of one. An exact rule of the evolution, that every arrival must eventually die, the well-adapted included, is what pushes the population onto that condition.

Public blockchains (often mistaken for cryptocurrencies) are, in a sense, an inevitable evolution of techno-social phenomena, and they will likely revolutionise key aspects of our social and economic life. Like many social phenomena, they emerge from a large number of people. Unlike most of them, they spread exponentially fast, thanks to the underlying technology: the internet, mathematics (hashing, zero-knowledge proofs and so on) and our ever-growing need for consensus and truth.

One aspect I have been looking into is how blockchain ecosystems organise themselves: the evolution of transaction networks, decentralised social networks, and the optimal construction of layer 2s like the Lightning Network.

On a more personal note, I am a (somewhat) active member of the blockchain community, as you might tell from my ENS name, fengling.eth. I take part in proof-of-work and proof-of-stake networks, vote in community governance, and report bugs.

Will decentralisation last, or will we keep cycling between decentralisation and centralisation (合久必分，分久必合), as human civilisations always have? This may be the first time in history that things are different.

Spreading events, from infectious disease over human contact networks to information over online social media, can switch between limited local spread and widespread global impact, much like a phase transition. I apply complexity theory to social media and other social networks to understand how such spreading works.

My work has looked at how influence travels beyond direct contacts, which we described as an induced percolation phenomenon; how the local structure of a network can identify and quantify its most influential spreaders; and how a mechanistic model of viral spreading on an empirical network can predict popularity. When people are overloaded with information, a message spreads unlike a disease and has a finite epidemic threshold. For epidemics, too little resource for treatment in the early stage makes the outbreak size jump abruptly.

The same tools describe how networks hold together. Community structure in interdependent networks can hide an extreme risk: weakening the communities can abruptly drive the whole system into a fragile state. Other work shows that how predictable a network's structure is can be read from how well it compresses, and recovers with machine learning the order in which a network's links formed. Current work maps the structure of the X (Twitter) network in Singapore.

IndependentShared strategiesPause

A financial market is made of many traders whose decisions interact, and its prices share universal statistical features: returns with fat tails, and volatility with long-term memory. With an agent-based model we showed that fat tails arise when traders share similar technical trading strategies. Extending the model to a stochastic one, we linked the long-term memory of returns to the way investors evaluate their investments over different time horizons, in agreement with empirical data.

Related work adds adaptive confidence to the traders, which raises price volatility; quantifies systemic risk in networks of banks and assets, where collective failure can change discontinuously; and follows the time-evolving network of venture-capital investment in China.

The movements of millions of people in a city add up to regular, law-like patterns of flow. I study these patterns with transit, flight and travel-survey data, mostly from Singapore.

With transit data from before, during and after the COVID-19 pandemic, we found that three classic theories of human mobility, the gravity model, the radiation model and the visitation law, all stayed valid, with the visitation law the best throughout and the gravity model improving during and after the pandemic. At the global scale, an airport's passenger volume follows its place in the flight network, best measured by PageRank weighted by the number of airlines; Asian airports carry more traffic than their connectivity predicts. For transport simulations, we built a generative model that creates synthetic itineraries for every private-vehicle user, keeping the statistics of the travel survey while protecting privacy.