My hometown has a genuinely strong orchestra — the Guangzhou Symphony Orchestra. It doesn’t have an all-star woodwind section like the Berliner Philharmonic, but having played in its affiliated Youth Orchestra during high school, I know the musicians are well-trained professionals. Plus, our artistic directors and conductor are not only well-connected but are themselves part of the core of China’s elite classical music scene. Oh, and fun fact: Guangzhou Symphony Orchestra often don’t give encores. ...

Tmux is powerful for parallelization but can be frustrating for monitoring experiments at scales. Like, you can run multiple parallel projects in different sessions but it’s hard to monitor each session — because you will have to manually ‘attach’ to a session to see outputs, and tmux has limited scrollback history and clunky scrolling interaction (you’ll have to Ctrl-v [ into copy mode to scroll). And once a tmux session dies — all terminal outputs are lost. ...

Came across a paper that proposed an interesting mechanism to improve rating systems. This is a very very good idea. RewardRating: A Mechanism Design Approach to Improve Rating Systems Vakilinia, Faizian, Khalili. Games (2022) https://doi.org/10.3390/g13040052 “To improve rating systems, in this paper, we take a novel mechanism-design approach to increase the cost of fake ratings while providing incentives for honest ratings.” … “Our proposed mechanism RewardRating is inspired by the stock market model in which users can invest in their ratings for services and receive a reward on the basis of future ratings.” ...

I was surprised to discover a lovely — and fairly well-loved — harp-flute duet that had somehow escaped my notice: The Fog is Lifting, Op. 41, by Carl Nielsen. It’s always a joy to stumble across a hidden gem in the repertoire. This piece isn’t about flashy technique or dense chromatic harmonies. What it does offer is space — space for the performers to shape tone, color, and atmosphere. It’s all about timbre. ...

Our Beyond-Bayesian-Bandits reading group covered this paper today: Multitasking: Efficient Optimal Planning for Bandit Superprocesses Dylan Hadfield-Menell and Stuart Russell. [Link](https://people.csail.mit.edu/dhm/files/bsp_bbvi.pdf to paper) and its supplementary materials. A bandit superprocess is a decision problem composed from multiple independent Markov decision processes (MDPs), coupled only by the constraint that, at each time step, the agent may act in only one of the MDPs. Multitasking problems of this kind are ubiquitous in the real world, yet very little is known about them from a computational viewpoint, beyond the observation that optimal policies for the superprocess may prescribe actions that would be suboptimal for an MDP considered in isolation. (This observation implies that many applications of sequential decision analysis in practice are technically incorrect, since the decision problem being solved is often part of a larger, unstated bandit superprocess.) The paper summarizes the state-of-the art in the theory of bandit superprocesses and contributes a novel upper bound on the global value function of a bandit superprocess, defined in terms of a direct relaxation of the arms. The bound is equivalent to an existing bound (the Whittle integral), but is defined constructively, as the value of a related multi-armed bandit. We provide a new method to compute this bound and derive the first practical algorithm to select optimal actions in bandit superprocesses. The algorithm operates by repeatedly establishing dominance relations between actions using upper and lower bounds on action values. Experiments indicate that the algorithm’s run-time compares very favorably to other possible algorithms designed for more general factored MDPs. ...

I was — and still am — a huge Sherlock Holmes fan. I first got into it around age 10, thanks to a Chinese edition I found during a summer holiday at my grandparents’ house. I barely remember the plots now, but the heat, the watermelon, and those long, lazy afternoons spent binging the stories without a single worry in the world were so memorable. There’s something about Holmes that makes him perfect summer reading. Today I randomly flipped to The Adventure of the Blue Carbuncle — easily one of my favorite short stories. I won’t spoil anything — but it has all the cleverness you’d expect from a Holmes case, plus a bit more fun and lighthearted with a cozy Christmas spirit baked in — like, it’s always a treat to watch Holmes playfully challenge Watson, like when he look at a hat and declares it a window into a man’s soul — while Watson just stares and baffled with the humor: ...

Incoherence is sufficient for exact matrix completion under random sampling. Coherent matrix can also be accurately recovered if random sampling can be contingent on each entry’s local coherence.

Why matrix completion is equivalent to minimizing the recovered matrix’s spectral norm.

Tchaikovsky didn’t like Wagner’s Ring of Nibelungen

(Ariana in Shanghai 2025)