Notes from 27/1–17/2

Hum Qing Ze

5 min readFeb 17, 2020

Golden Kitty awards from ProductHunt

I’d like to really take the time to think about what these products have that make them successful.

Social Impact Product of the Year: Wren

Hardware Product of the Year: MacBook Pro 16“

Mobile App of the Year: GitHub for Mobile

Design Tool of the Year: Figma Plugins

Dev Tool of the Year: Composer

Diversity and Inclusion Product of the Year: Elpha

⛓ Crypto Product of the Year: Snowball Money

No Code Product of the Year: Makerpad

Fintech Product of the Year: Predicto

AI & Machine Learning Product of the Year: FaceMaze

Side Project Product of the Year: Leave Me Alone

Remote Work Tool of the Year: Nomad List 5.0

WTF Product of the Year: Bird News Now

Audio & Voice Product of the Year: Descript Podcast Studio

AR Product of the Year: Wanna Kicks

Health & Fitness Product of the Year: Zenia

Smart Home Product of the Year: Lazy

Privacy Focused Product of the Year: Brave

SexTech Product of the Year: Coral

D2C Product of the Year: Haus

PetTech Product of the Year: Whistle GO

Product Memoriam: Inbox by Google

Product of the Year: AirPods Pro

8 tips for better negotiation — or, how to gain a magical superpower

1. Prepare, prepare, prepare

2. Start high, but not stratospheric

3. Prepare your BATNAs (best alternative to a negotiated agreement)

4. Listen

5. Remember negotiation is just one stage of a longer-term relationship

6. Don’t treat the other side as the enemy

7. Don’t focus purely on money

8. Post-Agreement Negotiation

What Viktor Frankl’s logotherapy can offer in the Anthropocene

Frankl writes that he ‘grasped the meaning of the greatest secret that human poetry and human thought and belief have to impart: The salvation of man is through love and in love.’ It is easy to be cynical about such a claim, proving Frankl’s point. In our small, petty, limited, cruel era, it seems hard to come across much collective human affection, and yet our pettiness, limitations and cruelty are in their own way a response to the looming apocalypse.

AI

NeurIPS 2019 Highlights

[New] Graph Neural Networks: Graphs were almost everywhere at this NeurIPS conference. Graphs are very powerful non-Euclidean data structures that can represent complex relationships and interdependency between objects. Hence data in a large number of applications (such as social networks, physical science, chemistry, e-commerce, knowledge bases, recommendation systems, and combinatorial problems) can be represented as graphs. The irregular structure of graphs (variable size and unordered neighbor sets for the nodes) makes it challenging to perform operations such as convolution on them. So there has been an increasing interest in Graph Neural Networks (GNNs) algorithms in order to generalize deep learning approaches such as CNNs, RNNs, and autoencoders for graph data. GNNs can be categorized into four groups: recurrent graph neural networks, convolutional graph neural networks (CGNNs), graph autoencoders, and spatial-temporal graph neural networks, with the heavier weight on CGNNs so far.

Comparison of a 2D convolution on the left, with graph convolution (neighbors of a node are unordered and variable in size) on the right [Image taken from this paper].

To get to know more about GNNs, take a look at the post A Gentle Introduction to Graph Neural Networks or the following survey papers:
• A Comprehensive Survey on Graph Neural Networks,
• Stanford SNAP’s Graph Neural Networks Tutorial, and
• Graph Neural Networks: A Review of Methods and Applications.
A good list of GNN papers is also collected in this repo. Graphs not only can be used as a data structure, but also can represent outputs of NNs, e.g. any joint distribution can be represented as a factor graph as noted by Yoshua Bengio in his invited talk.

Graph representation learning

Blockchain

What is Blockchain and how does it work?

Keeping it simple, nice explanation

Blockchain Ethereum

How to Write Smart Contracts That Optimize Gas Spent on Ethereum

List of patterns for solidity

Build an Ethereum dapp using ethers.js

Building a typical voting dapp as an example

What Comprises an Ethereum Fullnode Implementation?

Comparison between full node and archive node

But, as Ethereum continues to mature, and is embedded in more devices, its flexibility will begin to shine and continue to improve (with a fresh blank state with the incoming transition to ETH 2.0, making things even more efficient storage-wise).
We see a future of many clients, all acting as independent economic agents, with different data availability requirements. But, regardless, a fullnode on Ethereum is in fact that, a fullnode. And, how one syncs theirs should not be misconstrued with other chains like Bitcoin, as long as you’re using the right type of node for the job.

Here’s each of the various types of clients you can configure on Ethereum:

Light Clients: Requires no validation, requests the current state from the P2P network to verify current state (fine for processing payments and simple contract calls), but your validation is out-sourced to other fullnodes with the necessary information.
Fast Node (Fast Sync/Warp Sync): Will not validate intermediate states during the initial sync, will validate everything else after that, however. This allows older data to be pruned that likely has nothing to do with your transactions
Full Node: As described above, validates everything, and will prune old intermediate states from memory, and will keep an archive of future state trie updates.
Archive Node (Historical Node): Validates and stores all intermediate states, nothing is pruned, all state transitions for accounts are retrievable. Future state trie updates and full intermediate states are stored.

Ethereum Hackathon Survival Guide

You get redirected to the Consensys Dev portal.

Seems that there is a real need to set up some sort of Hackathon preparation checklist?

A Deep Dive Into The Ethereum Virtual Machine (EVM)

Stumbled onto Kauri, it’s really a couple of people writing articles that… quote the paper? Most interesting

Data

I can’t teach you Data Science in 10 days

I believe a Data Analyst plays the most crucial role in any Decision Cycle chain. Two reasons. “One, they are the key lever towards ensuring data is procured, transformed and stored in a way that it is structured and ready to be used. Two, most of the organizations have huge volumes of data that require time and skill to be processed into a useful form”.

Bottom line is there are different types of analysts that serve different aspects of the business

Fooled By Randomness: Applying Monte Carlo Simulations to Fund Managers

Now the fund manager analogy isn’t really a useful one if you’re just assigning people random chances of survivability. It simply shows that when you randomly generate things, some live and some die.

Development

Mint

The programming language for writing single page applications.

Mint has all the tools you need to write error free, easily readable and maintainable applications in record time.

One Hundred Ideas for Computing

Absolutely brilliant! These ideas are interesting in the sense that they don’t seem to be done completely yet but are reachable (buildable)