Archive for Tokyo
Kazuo Iwamura (1939-2024)
Posted in Books, Kids with tags Chevalier des Arts et Lettres, children books, Japan, Japanese literature, Kazuo Iwamura, kids, L’École des loisirs, mice, obituary, Tokyo, translation, 岩村 和朗 on January 30, 2025 by xi'an
a journal of the chaos, conquest, war, [stale nuts] famine, and death year
Posted in Books, Kids, Mountains, pictures, Running, Travel with tags book review, Chamonix-Mont-Blanc, COVID-19, Daniel Defoe, film review, French cheese, hazelnut, Hiroshige, Journal of the Plague Year, Judge Dee, Kyoto, Morocco, msemen, Netflix, nut butter, onions, pizza, Ramadan, Robert van Gulik, Russian invasion, teen culture, Tokyo, torrefaction, Tōkaidō, Ukraine, Utagawa, woodcuts on May 4, 2024 by xi'an
Read The Big Wall (in French), a manga by Yoji Kamata & Kunihiko Yokomizo, which my wife bought me while visiting the mountaineering Éditions Guérin in Chamonix. This is in fact a collection of seven stories about climbing and mountaineering, in Japan, the Himalayas, and Canada, all related with the same central climber. It is a bit similar to The Summit of the Gods, another if much earlier gift from my wife!, in that an outstanding climber, Yasushi, faces extreme situations, saves lesser climbers’ live while trying to solve a personal dilemma. The similarity extends to the highly unrealistic climbing situations where Yasushi survives despite all odds! I also finished Pascale Quiviger’s Royaume de Pierre d’Angle (3), again in French, with many more days and less enthusiasm than the earlier volumes as the story pace drags down to a confrontation between both royal brothers, with too many Deus ex Machina moments and the sad erosion of the ethics of 
Thibault, the current (and reluctant) king. I also started reading (the French translation) of Along the Tōkaidō trail, (Tōkaidōchū Hizakurige, 東海道中膝栗毛), by Jippensha Ikkû, which is a 1802 touring guide of the road between Tōkyō (Edo) and Kyōto, and its famous 53 stations celebrated by Utagawa Hiroshige in famous woodcuts. Its main appeal is that the guide is a pretext for a ribald novel about two rascals walking the road and the depiction of the Japan life and popular culture in the early days of the Edo Period. However, it is neither a bedside nor a travel book, since I need permanent access to internet to locate place, check the corresponding HIroshige woodprints, and read through the numerous footnotes explaining puns (like its title translating as Shank’s Mare) or special traditions, therefore travelling quite slowly on the Tōkaidō trail!
Cooked most days upon returning from the Alps, if nothing particularly novel, except for pissaladières made with Moroccan msemen basis rather than foccacia bread, since they were sold in the local bakery during Ramadan (along with Moroccan pastillas that were in high demand!). Making these “Nice pizzas” a quick dinner option, for only requiring frying onions for a few minutes and adding anchovies. I also made a few jars of hazelnut butter, upon advice from a visiting friend for using the bags of hazelnuts collected in late summer by my mother a few years ago! They had predictably turned sour but enough torrefaction hid the sourness to make the butter enjoyable (after a few attempts).
Watched a Netflix attempt at Judge Dee stories, which I found lacking when comparing with the original books by Robert van Gulik, especially since it is very loosely connected with some van Gulik’s stories (like The Chinese Bell Murders, The Chinese Gold Murders, The Haunted Monastery, The Lacquer Screen, The Red Pavilion). In addition, the underlying threat from a mysterious revolutionary organisation ruins the purpose of the story and the sleuthing abilities of Judge Dee often appear secondary. Some aspects remain appealing, like in the story centred on a puppet theatre show or the one where the central imperial authority gets questioned by the population of a Gobi town. Overall, I also disliked highly unrealistic resolutions (like the one involving a parting of the sea that was even more unbelievable than in the Ten Commandments!) and the very slow progress of Dee’s relationship with a courtesan whose past life is deeply intertwined with Dee’s family history (and missing from the books).
a journal of the conquest, war, famine, and death [definitely new] year
Posted in Books, Kids, Mountains, pictures, Running, Travel with tags Aardman Animation, anime, Aubrac, BD, Blake and Mortimer, book review, books, Bristol, buckwheat, cheese, consumer prices, COVID-19, Daniel Defoe, Edgar P. Jacobs, Edo era, film review, French cheese, galette, Japan, Journal of the Plague Year, oysters, red cabbage, Russian invasion, Saint-Chély d'Apcher, Saint-Chély d'Aubrac, samurai, teen culture, The Art of War, Tokugawa, Tokyo, Ukraine, warhammer, Yule on February 7, 2024 by xi'anRead some graphical novels over the Yule break, incl. the new volume in the endless Thorgal series, Adieu Aricia, created by a master of heroic fantasy comics, Recht Robin, L’Art de la guerre, a refreshing if light pastiche of Blake & Mortimer drawn by Floc’h, far from the E.P. Jacob ligne claire (following another one by another master of Belgian comics, Schuiten, Le Dernier Pharaon), and Le Monde sans fin, by Blain & Jancovici, a best seller retracing the central importance of energy in evolution and the urgent need to change the entire societal software (if a bit heavy handed).
Cooked little while being away, except for a Flemish red cabbage stew and a batch of mini buckwheat galettes, and fortunately escaped the huge meals that usually plague this time of year. Avoided as well some viral issues with French oysters, despite enjoying several servings of Norman origin. While in Aubrac, has a taste of the local beef, the well-named… Aubrac breed, whose herds are raised on this immense and treeless plateau during the warmer months. Most crucially, we enjoyed very much the few days we spent there in a remote farm with no Internet, great company, long trail runs, and simple home-made dinners. (Foodwise, the difference of food prices on the local markets like Saint-Chély d’Apcher when compared with Paris and its suburb was noticeable, if not a major surprise, meaning we brought back home a month worth of local cheese.)
Watched Chicken Run #2 with my kids, which I found heavy handed and predictable, even on a 25 December afternoon, if admirable stop motion animation by the Bristol based Aardman studios, Rebel Moon by myself, just a terrible mix of existing space operas, with frankly embarrassing moments (like the disguised horses, the Warhammer-ish fights, and the final scene when the band is carelessly crushing through a wheat field) and Blue Eye Samurai with the whole family. The latter is a French-American anime set in Japan, around the Great Furisode Fire of 1657 in Edo and the expulsion of foreigners from Japan by the Tokugawa shogunate. Graphically beautiful, Great except for the dialogues being in English, the cubist way horses are drawn and for some historical inaccuracy.
Bayesian learning
Posted in Statistics with tags Bayesian learning, Herman Rubin, mixtures of distributions, Newton-Raphson algorithm, One World ABC Seminar, RIKEN, stochastic gradient descent, Tokyo, webinar on May 4, 2023 by xi'an“…many well-known learning-algorithms, such as those used in optimization, deep learning, and machine learning in general, can now be derived directly following the above scheme using a single algorithm”
The One World ABC webinar today was delivered by Emtiyaz Khan (RIKEN), about the Bayesian Learning Rule, following Khan and Rue 2021 arXival on Bayesian learning. (It had a great intro featuring a video of the speaker’s daughter learning about the purpose of a ukulele in her first year!) The paper argues about a Bayesian interpretation/version of gradient descent algorithms, starting with Zellner’s (1988, the year I first met him!) identity that the posterior is solution to
![\min_q \mathbb E_q[\ell(\theta,x)] + KL(q||\pi)](https://s0.wp.com/latex.php?latex=%5Cmin_q+%5Cmathbb+E_q%5B%5Cell%28%5Ctheta%2Cx%29%5D+%2B+KL%28q%7C%7C%5Cpi%29&bg=000000&fg=B0B0B0&s=0&c=20201002)
when ℓ is the likelihood and π the prior. This identity can be generalised to an arbitrary loss function (also dependent on the data) replacing the likelihood and considered for a posterior chosen within an exponential family just as variational Bayes. Ending up with a posterior adapted to this target (in the KL sense). The optimal hyperparameter or pseudo-hyperparameter of this approximation can be recovered by some gradient algorithm, recovering as well stochastic gradient and Newton’s methods. While constructing a prior out of a loss function would have pleased the late Herman Rubin, this is not the case, but rater an approach to deriving a generalised Bayes distribution within a parametric family, including mixtures of Gaussians. At some point in the talk, the uncertainty endemic to the Bayesian approach seeped back into the picture, but since most of the intuition came from machine learning, I was somewhat lost at the nature of this uncertainty.
the Bayesian learning rule [One World ABC’minar, 27 April]
Posted in Books, Statistics, University life with tags ABC, Approximate Bayesian computation, approximate Bayesian inference, Bayesian deep learning, Bayesian learning, Kalman filter, Newton-Raphson algorithm, One World ABC Seminar, RIKEN, stochastic gradient descent, Tokyo, University of Warwick on April 24, 2023 by xi'an
The next One World ABC seminar is taking place (on-line, requiring pre-registration) on 27 April, 9:30am UK time, with Mohammad Emtiyaz Khan (RIKEN-AIP, Tokyo) speaking about the Bayesian learning rule:
We show that many machine-learning algorithms are specific instances of a single algorithm called the Bayesian learning rule. The rule, derived from Bayesian principles, yields a wide-range of algorithms from fields such as optimization, deep learning, and graphical models. This includes classical algorithms such as ridge regression, Newton’s method, and Kalman filter, as well as modern deep-learning algorithms such as stochastic-gradient descent, RMSprop, and Dropout. The key idea in deriving such algorithms is to approximate the posterior using candidate distributions estimated by using natural gradients. Different candidate distributions result in different algorithms and further approximations to natural gradients give rise to variants of those algorithms. Our work not only unifies, generalizes, and improves existing algorithms, but also helps us design new ones.
