Equations for the kitchen under the Christmas tree

Magical constants

C — Cleanliness (see Marshak),

B — Balance (see Nash),

MLPP — Materials, people, and process in space (see Mayakovsky, Corbusier).

A fresh graph convolutional neural network is just as useful in the kitchen. Playing molecular cuisine has never been easier. It's enough to describe the dish as:

$inline$H = sigma [ hat D ^{-½} hat A hat D^{-½} Dropout {sigma[hat D ^{-½} hat A hat D ^{-½} X W ]}] , $inline$

Where:

Equations for the kitchen under the Christmas tree — regularization by zeroing randomly chosen parameters — that ‘forgetting’ patented by Google,
— nonlinear activation function ,
— diagonal matrix of vertex degrees (number of connections),
— diagonal matrix of ones,
— connectivity matrix,
Equations for the kitchen under the Christmas tree — vertex properties matrix,
— weights of the neural network that we can:

learn in supervised learning mode;
fill with randomly chosen values;
discard (the function will become slightly more linear).

The author examined the mechanics in detail himself here. Furthermore, the method is also explored step-by-step with examples here.

About how to get into graph embeddings — an introduction to graph theory and the corresponding notation can be refreshed — I’ve written about it before.

Now — to what’s truly important. On preparing data labels.

For five years at Nestle, I learned how specially trained people and robots made chocolates from raw materials and packaging. My first encounter with the heart of the business occurred early in the spring when I implemented product data lifecycle processes based on SAP — from idea to market exit. Now — I’m observing the process of preparing bachelor chefs in Lisbon. In the picture above — one of the boundaries in pastries — a candy with egg cream. Like a Christmas ornament made of sugar, only stuffed. Here, they train chefs over four years.

I am thrilled by what I saw. And since I am sharing educational materials here, I will explain why it’s better to study cuisine in Portugal than mathematics.

Equations for the kitchen under the Christmas tree

Hogwarts welcomes with a bas-relief inscribed 'Labor and Order'.

This is precisely what students can expect. By chance, I was educated in both safety and environmental protection in food production — I managed market certification (a legal entity representing a regional unit from two countries — a couple of factories, an office, and a couple of warehouses) according to all relevant ISO standards and developed internal audits for a global market leader — an interesting experience. It provides a basis for evaluative judgment.

Located in the trendiest area of Lisbon, the school of tourism and hospitality is a bastion of the right approach to business.

Equations for the kitchen under the Christmas tree

What is the point of paying attention to zucchini salad on Habr — you might ask?

The point is that there are eternal values, like pastries and clever solutions, while code is transient. For example, in a couple of days, a second Python will already capture the state of development. Increasingly, conversations about ethics in applying mathematics are becoming more frequent, and I believe that pastries provide an excellent field for experiments in molecular gastronomy.

One of the discoveries was chocolate mousse with strawberry marmalade. Black caviar with white chocolate is becoming a cliché, yet it remains an impressive combination. Overall, it's fun, sometimes delicious, ethical, and new.

Is it possible to combine mathematics with cooking in Lisbon?

Most likely, yes. In self-study mode.

What else is interesting to incorporate into your food processor?

Here are two recommendations — a recently presented strategy for pre-training graph convolutions and a new distance measure for generative models, which will only be presented at ICLR 2020.

Equations for the kitchen under the Christmas tree

The first is exciting because it enables learning the fine details of ingredient relationships on publicly available data and fine-tuning to local specifics — this data will need to be collected, studied, prepared, etc. — much like developing a restaurant menu or a technological chart for individual dishes. We discussed this work at the recent NeurIPS Lisbon Meetup — organized by PyData-Lisbon in a format over two consecutive days from 2 PM to 8 PM — thanks to them for regularly delivering talks, pastries, and pizza dropouts.

Equations for the kitchen under the Christmas tree

It's a beautiful field. It is still unclear how much this distance measure will accelerate progress in creating recipes and technological charts from photographs, but the potential is there.

In brief, a graph can approximate a manifold. Its shape can be quite accurately learned by modeling the propagation of heat within the structure.

The mathematics involved can be daunting at times, and in general, graph theory is rapidly growing with tools and the selection of algebras is expanding — there is more talk about hypergeometry and other spaces. More importantly, the application specifics — after all, a furnace can also be viewed as a black box.

As a result, we can more effectively evaluate the recipes generated by the neural network.

It is well known that training neural networks is a complex engineering discipline. It consumes a lot of data, electricity, and machine hours. So if you haven't digitized grandma's cookbook to train refined architectures for throttle rumbling, and you're still skeptically glancing at algorithmic cooking, there is an alternative. People in white hats.

If you are not yet ready to hand kitchen work over to robots but are opening a restaurant, I highly recommend considering graduates from this establishment. They are well trained there.

Equations for the kitchen under the Christmas tree