Startup Blogs

Flat designt business startup launch concept.

How to start a startup

Startup

Founders at Work

The Best Simple Business Models

Why I’m leaving silicon valley

未来世界的幸存者

Software and technology stack used by top companies

Do things that don’t scale

 

Donate $5 to me for a coffee with PayPal and read more professional and interesting technical blog articles about web and mobile development. Feel free to visit my web app, WhizWallet, to apply for credit, store or gift cards, DealsPlus to browse daily deals and store coupons to save money, BlackFriday.fm to check latest news, ads and sales in BlackFriday shopping season.

Follow me @Yaoli0615 at Twitter to get latest tech updates.

Advertisements
Posted in IT, startup, Uncategorized | Tagged , , | Leave a comment

Python Blogs

567828_67d0

Google’s Python Class

Dive into Python

A Python Reading List

Map, Filter, Reduce

Python Tutorial: Shallow and Deep Copy

Python Numpy Tutorial

PEP 8 — Style Guide for Python Code

Python Zen – FooFish Python之禅

 

 

Donate $5 to me for a coffee with PayPal and read more professional and interesting technical blog articles about web and mobile development. Feel free to visit my web app, WhizWallet, to apply for credit, store or gift cards, DealsPlus to browse daily deals and store coupons to save money, BlackFriday.fm to check latest news, ads and sales in BlackFriday shopping season.

Follow me @Yaoli0615 at Twitter to get latest tech updates.

Posted in CS Research&Application, Uncategorized | Tagged | Leave a comment

Kwik Learning

20065291_791032531070507_9203537912241586176_n

1. Manage first hour and last hour.

 

2. 10 things to do in the morning:

1) remember your dream (i.e. business, ML/AI, 3 commas)

2) make your bed

3) drink lots of water

4) brush teeth with opposite hand

5) deep breath exercise (meditation)

6) drink brain tea

7) journaling (record daily plan and reflection)

8) 5 minute HIIT workout (to warm up)

9) drink brain power smoothie (blueberry, avocado, etc)

10) daily reading

 

3. Learn FAST

Forget: be hungry, be foolish

Active: take notes, concentrate

State: emotion influence learning effect

Teach: learn by teaching

 

4. Self talk is very important.

“As you think, so shall you become.” – Bruce Lee

 

5. “Education is ultimately much more than simply memorizing individual facts, or even learning individual concepts. [What] matters most: learning how to think, learning how to reason and learning how to learn.” – Vitalik Buterin

 

6. Ask questions, be dump for seconds, then wise ever after

 

Donate $5 to me for a coffee with PayPal and read more professional and interesting technical blog articles about web and mobile development. Feel free to visit my web app, WhizWallet, to apply for credit, store or gift cards, DealsPlus to browse daily deals and store coupons to save money, BlackFriday.fm to check latest news, ads and sales in BlackFriday shopping season.

Follow me @Yaoli0615 at Twitter to get latest tech updates.

Posted in Book/Movie/Music, Uncategorized | Tagged , , | Leave a comment

Favorite Chrome Extensions

best-chrome-extensions-670x335

Momentum: wallpaper, todo list, quotes

SimilarWeb: similar websites

Wappalyzer: website technology analysis

Google Optimize: A/B testing

AdBlock: block ads

 

Donate $5 to me for a coffee with PayPal and read more professional and interesting technical blog articles about web and mobile development. Feel free to visit my web app, WhizWallet, to apply for credit, store or gift cards, DealsPlus to browse daily deals and store coupons to save money, BlackFriday.fm to check latest news, ads and sales in BlackFriday shopping season.

Follow me @Yaoli0615 at Twitter to get latest tech updates.

Posted in CS Research&Application, Uncategorized | Tagged , | 2 Comments

Block Chain, Bitcoin, Ethereum Notes

shutterstock_554011084

Users use private key to manage address and transfer bitcoin need verify with all nodes.

Anonymous vs Pseudonymous (BitCoin, public chain)

Bitcoin miner income includes new added bitcoin and transaction fee, miners record transaction and gain bitcoins.

Mining needs to compute random number (nonce) and it likes gambling.

Block Chain is a kind of distributed System which solved the Byzantine General Problem.

BitCoin bring economy idea to algorithm, giving money to nodes to prevent people do bad things.

BitCoin satisfy AP in CAP theorem and will be consistent eventually.

Computing power is larger than 51% network and it’s absolute unsafe, less than 45% is relative safe.

POW (proof of work, computing power) vs POS (proof of stake, money)

Wet code is for human and dry code is for machine.

Block Chain works as database (manage address) but also have computing power (record transaction data).

Bitcoin script: not Turning complete, have no loop statement.

Smart Contract: public trust network

Ethereum: contain loop, almost Turing complete, a platform to use Block Chain to build applications.

White paper like business plan, yellow paper like descriptive documentation.

Use gas to calculate (add, subtract), which will break infinite (“dead”) loop.

Programming: Ethereum Virtual Machine (EVM), Solidity (JS like), Serpent (Python like)

Block Chain: A trusted computing machine and database. Another weapon works as a decentralized server and transfer value. (contract is like a class in Solidity)

Parallel write: clients make request (JSON RPC call) to consensus node which will decide to execute which request.

different price contracts: 99, 100, 123, 156, 567, 897

Application fields:

Finance: data correction (two banks have different data)

IoT: share economy (rent, lease, share lock with smart contract), supply chain, logistic, patent and copyright (music, art) management.

Two issues for Block Chain: privacy and scalability.

Homomorphic encryption to solve privacy issue in public chain, or use private chain.

Block Chain is a kind of replicated state machine, there are applications like: Alarm Clock (execute scheduled task), RANDAO (generate random number).

IPFS and Swarm to resolve storage issue.

There is a trend to overuse economic encouragement, it should try to figure out in algorithm perspective.

 

Donate $5 to me for a coffee with PayPal and read more professional and interesting technical blog articles about web and mobile development. Feel free to visit my web app, WhizWallet, to apply for credit, store or gift cards, DealsPlus to browse daily deals and store coupons to save money, BlackFriday.fm to check latest news, ads and sales in BlackFriday shopping season.

Follow me @Yaoli0615 at Twitter to get latest tech updates.

 

Resources:

BitCoin and Block Chain

Ethereum

How the Byzantine General Sacked the Castle: A Look Into Blockchain

How does Bitcoin force consensus among Byzantine generals?

How does Ethereum work, anyway?

Posted in CS Research&Application, IT, Uncategorized | Tagged , , | Leave a comment

Deep Learning Notes – Activation Function

500_f_84517548_t640subkwc91b36ek9yzypvwsibb3nvn

Activation function

Sigmoid: Vanishing gradient (but still used in RNN. gradient for multi-D and slope for 1-D)

ReLU:sub differential (corner point could be derivative)

Tanh: (range from negatives)

 

Train model and learn parameters to minimize loss

 

Loss function

Cross Entropy (predict categorical labels, K.L. distance between ground truth and prediction, sum of ground truth and log prediction, could be very steep and easier to update parameter to find minimum loss)

Squared Error (predict continuous values, sum squares of ground truth and prediction, not steep as Cross Entropy for categorical labels)

 

Stochastic Gradient Descent (SGD)

Check partial samples, prematurely update model, maybe bring randomness, but fast to converge. (Initial weights: Random or RBM)

 

E.g. y=2x^2, assume initial value is 4, then update w’ = w – ita * delta (ita as learning rate and delta as gradient) = 4 – 0.1 * (4 * 4) = 2.4

 

AdaGrad

Smaller Derivatives, Larger Learning Rate

If learning rate is too large, it might be converged

 

Avoid overfitting

Dropout (only in Deep Learning)

Training (neuron with p% to turn off, balanced with neurons turned on, whose weights will be updated to larger number)

Testing (no drop out, but all weights times 1-p%)

 

Ensemble model (with average vs Random Forest with majority vote)

Layers close to input is like use sub-feature space

Dropout rate is usually less than 0.5 in production (0.1 to 0.3 most time, layers close to input could be higher and output ones lower to keep more learned information for final result)

Higher Dropout rate, longer training time

 

Donate $5 to me for a coffee with PayPal and read more professional and interesting technical blog articles about web and mobile development. Feel free to visit my web app, WhizWallet, to apply for credit, store or gift cards, DealsPlus to browse daily deals and store coupons to save money, BlackFriday.fm to check latest news, ads and sales in BlackFriday shopping season.

Follow me @Yaoli0615 at Twitter to get latest tech updates.

 

Resources:

Pattern Recognition and Machine Learning

Machine Learning with R

Machine Learning Map

Machine Learning Algorithms Pros and Cons

Mathematical Decision Making: Predictive Models and Optimization

Posted in Artificial Intelligence, Uncategorized | Tagged , | Leave a comment

AI Serving

ai-slideshow-solve-problemsa-712

1. Serving

End to end system (given input and provide output, accept request and give response)

AI serving (using AI technique to provide response for request)

 

2. Traditional Web Application

Whole page reload when direct to other pages

 

3. Single Page Application (SPA)

Browser does most of heavy lifting, make AJAX call to request JSON result

Fast, easy to debug (keep logs until reload), friendly for mobile apps, cache local storage

 

4. Memory leak

C++: forget to release declared resource (char *ptr = new char[10]; dele[] ptr; )

c# / java: class member variable holds native resource (hardware related, file handler, socket) which need to be garbage collected when destruct class instance, otherwise memory leak.

node.js : node.js or browser bug

Python: python internal bug

 

Donate $5 to me for a coffee with PayPal and read more professional and interesting technical blog articles about web and mobile development. Feel free to visit my web app, WhizWallet, to apply for credit, store or gift cards, DealsPlus to browse daily deals and store coupons to save money, BlackFriday.fm to check latest news, ads and sales in BlackFriday shopping season.

Follow me @Yaoli0615 at Twitter to get latest tech updates.

 

Resources:

Pattern Recognition and Machine Learning

Machine Learning with R

Machine Learning Map

Machine Learning Algorithms Pros and Cons

Mathematical Decision Making: Predictive Models and Optimization

Posted in Artificial Intelligence, Uncategorized | Tagged , | Leave a comment