The problem statement was a part of TECHGIG's hackathon held from 18th July to 2nd Sepember 2018. The hackathon Springer Nature Pune Hack Day: Decoding Scientific Content – Hosted by Springer Nature Pune on decoding scientific content, is a challenge designed to inspire creative and dynamic technology professionals to put their skills to the test. Spread over four weeks, Springer Nature Pune Hack Day: DSC is a platform that allows aspiring tech experts to display their technical prowess in new technologies including ML, NLP, AI.
Every publisher aspires to automate the identification, classification, enhancing, etc. of images using computer vision, which has a large variety of practical applications. Since this task of recognizing a visual concept is relatively trivial for a human to perform and is worth considering the challenges involved are from the perspective of a computer vision model. This problem statement talks about the need to automate the identification of the images as color or grey, which would help the publisher in decision making.
Background Information
The manuscripts submitted by authors have many images inside it. These images can be either colour or grey or both. As part of the business services, the author submitted images will be processed by the publisher before publication. The processing of images differs for color and grey scale images. The identification of this helps to further automate their processing.
Use Case Description - What we have tried to build
The Image Colourfulness Identification system
The system to identify whether the image is grey or colour (with good sensitivity)
Use Cases:
Use Case #1: Verify whether the image is grey or colour along with the percentage of confidence
Use Case #2: Must identify an image is colour even if the respective image has a small, clear and distinct colour information on it
Use Case #3: Must identify an image is grey even if the respective image has a very extremely small degree of colour artefacts in it (not noticeable visually but will/may be computationally)
It automates the identification of the images as colour or grey, which would help the publisher in decision making
Problem Being Solved
Avoid the manual verification of the images as colour or grey
Fully automate the current partial automated process to identify images as colour or grey
Reduce the probability of image misclassification
Reduce the additional cost that an author has to bear due to faulty processing of images
Avoid an additional manual visual checking of each image which would also save the time
Help the publisher in decision making
Help to further automate publishers processing
Classify images with good percentage of confidence
Technology/Tool/Cloud Stack
Machine Learning, Deep Learning
Convolutional Neural Network (CNN)
TensorFlow
Python
OpenCV
Softmax
Shape function
Hardware Specifications
CPUs
GPUs
Solution Approach and Architecture
Demonstration Video/Prototype Demo
Why This Solution Should be Considered
Platform Independent - Runs on verity of platforms - from desktops to clusters of servers to mobile and edge devices
Easy deployment - can run on multiple CPUs, GPUs and TPUs (Tensor processing unit)
System identifies an image is colour or grey along with confidence factor
System identifies grey images with extremely small colour artefacts also as grey with good percentage of confidence
System identifies even images with small, distinct/clear colour information on it as colour with good percentage of confidence
Solution is based on Training Data, Validation Data and Test Data
Challenges Faced
Selecting the solution approach
Deciding the solution model which fits the requirements
Possible Improvements
Select better network design to get more accurate output\predictions
Develop a truly Decentralized, Democratized, Transparent Micro-Lending platform based on Block chain. This platform would have to do the following:
Free market decision of interest rates which increases competition amongst lenders.
All Transactions should be visible transparently to everyone.
No middleman should be required just the borrower and lender.
Anyone having access to this platform should be able to borrow from remotest of village.
Platform should not provide limits on finance and duration of finance.
Based on past transactions it should build the reputations of borrowers and visible to lenders.
Abstract
Let us think about financial inclusion and digital inclusion and to provide financial lending services to people from low income groups who may be dire need of financial support for their livelihood. A typical example is that of a short-term repeat borrower in a city such as a trader who borrows money from a lender, buys vegetables from one or more farmers in the morning, sells the vegetables for a higher margin profit, and then returns the money to the lender at the end of the day. Borrowers could also borrow for longer durations such as a farmer that needs money to buy seeds and fertilizers to grow crops in a village.
Platform Description
The solution built upon Ethereum Smart Contracts
This Peer-To-Peer model – no need of any middleman to operate the system
Provides the financial lending services to the small group of borrowers
Provides ability to use existing networks to help build and verify borrower identities
Loans are facilitated through two models - one can be used at a time:
Model #1: Open Auction Model
Model #2: Random Number Model – With flat interest rate (not more than 2%)
Flexibility of choosing suitable model makes this platform unique
Problem Being Solved
Eliminates the lack of Trust and Transparency
Eliminates the existence of middleman between borrowers and lenders
Decision making freedom resides with borrowers and lenders
No mortgage/collaterals required while giving out loans to borrowers
Decentralised System
Security - The system allows borrowers to conduct deals directly which avoids the risk
Provides flexibility of choosing suitable solution model as per need/convenience
Enables crowd-sourcing anti-corruption measures
Technology/Tool/Cloud Stack
Ethereum smart contracts – Solidity
Ropsten TestNet
Truffle framework
MetaMask
Remix IDE
Steps to Compile
truffle init
truffle.cmd compile
truffle.cmd test
npm install truffle-hdwallet-provider
truffle.cmd migrate --network ropsten
Use 'remix-ide' command to test the contracts
Platform Architecture – Open Auction
The Open Auction Approach - Trust And Transparency
This is the process where certain people (e.g. from village) forms a group, we call it as a network.
Everyone decides a certain amount which is to be paid by everyone in the number of installments which is equal to the member count in the network.
The installment amount is fixed which is predetermined by the group.
The group hosts the series of open auction process on each month with a pre-scheduled date. Each group member has to deposit their installments monthly.
This open auction process continues for the number of consecutive months which is equal to the total number of members minus one (n-1) in the group/network.
In the auction process, each member can request for the loan from the network.
If the loan request is from only the single person, then the loan amount is to be disbursed by the system without any interest rate.
But, if the loan request is from more than one member of the group/network, then the one of them can take the loan by bidding the maximum interest rate amount.
The loan amount is fixed which is equal to the sum of all individual installments in one complete cycle of this process.
Once the loan is disbursed, the borrower is not allowed to take the loan again from the network in that cycle.
The system processes the loan to the winner who has put the bid with the maximum interest rate or to the eligible person who only has requested the loan amount.
The amount of interest will be pre-deducted from the loan amount itself and it will get equally distributed to each member of the network in that month itself.
The loan borrowers benefits from plenty of time to make profit from their funded business to arrange any dues and/or the installments to be paid in the complete cycle.
To avoid the people who are seeking this as an investment opportunity, the interest amount from the further installments(n-x) will not get distributed to the last x members of the network who are yet to take the loan from the system.
Instead, the interest amount will only get distributed to the remaining members of the group/network. This way, everyone in the network will get benefited equally during the cycle of installments.
The x factor is pre-determined by network and every member should agree to this decision.
The new member is allowed to join/leave the group only after completion of the current cycle.
e.g. 12 people come to form their own network. Everyone decides to deposit 10000/- as a monthly installments for 12 months. so, everyone is contributing their own amount equal to 120000/- in one complete cycle. Everybody is eligible to get 120000/- as a loan from the network once in the complete cycle starting from the first open auction. Everyone decides 4 as a x factor in the system. That means, in the last 4 installment cycles, whatever the interest rate that network earns from the auction process, it will only get distributed to the remaining people in the network those have already taken the loan. Suppose there are 3 eligible people who are requesting for the loan. then they have to win the loan amount by bidding it with maximum interest rate that they can afford. Consider a case where one has requested a loan by bidding the maximum interest rate equal to 5%. The system will disburse the loan amount (e.g. 120000 - 5% of interest) to the borrower who has won the loan request. The system will distribute the earned interest amount from this auction process depending upon the x factor and the number of current installments in the cycle completed.
Platform Architecture – Random Number
The Random Number Approach - Trust And Transparency
This approach is similar to the Open Auction Approach but with following differences.
Instead of requesting for the loan, it gets distributed to one of the random member in the network.
On every months scheduled date, the system generates the random number out of the number from the number = (total members in the network - # of people got the loan earlier in current cycle).
Additional to installment amount, everyone has to pay a pre-defined flat interest amount (e.g. 2%) to the network.
The collected interest amount will then be redistributed to the people in the network who has not yet got the chance to take the loan in current cycle.
This way, it will benefit to all in the network.
In case, if the person who has got the change to take the loan, but he/she doesn't want the loan to be taken at that time, he/she can offer the loan to the member in the network who has not yet got the chance to take it.
Advantages
Increases competition
Eliminates collusion risks almost completely
More secure
Increases transparency significantly
Enables crowd-sourcing anti-corruption measures
Increases trust among all the stakeholders
Disadvantages
One can't leave the network in-between
There is no provision of restructuring the installment amount/period if one wants to join/leave the network in-between.
Demonstration Video/Prototype Demo
Challenges Faced
Deciding the solution model which fits the requirements
Possible Improvements
Maintain Borrower’s/Lender’s Identity - e.g. linking Aadhar card to their digital identity
A simple User Interface is required
Improve Scalability – Can be scaled by allowing it to reach the greatest number of people around the world
Extra features can be added e.g. allow one to join/leave the network in-between the cycle
Provisioning reconstruction of instalment amount/period if one wants to join/leave the network in-between
