Someone who has always been interested in learning about new tools and technologies, Blockchain was something that was on top of the list. That’s why when I heard about the Celo India Fellowship, I knew I had to participate in it to get introduced to Blockchain Development and Decentralized Finance. For those who don’t know — Celo is a mobile-first blockchain platform that makes financial dApps and crypto payments accessible to anyone with a mobile phone. In this blog, I’ll go over every detail of my experience, from the application process to the completion of the fellowship.

Application Process

I got to know about the Celo India Fellowship from Devfolio — India’s largest and fastest-growing community of builders. For the application, long essay type questions were asked, which included one standout project of yours, any project ideas on blockchain, and why do you want to apply to the fellowship? After the initial application was submitted, an interview of 15 minutes was scheduled with the Devfolio community to see if I was a good fit for the fellowship. The interview was about how much previous knowledge I have of blockchain, some questions and terms related to it, the tech stack I have worked on, any project ideas that I have in mind, etc. Despite the fact that I was a complete beginner in blockchain, researching the topic helped demonstrate my interest in working on it and came in handy during the interview. After the interview was over, I got my letter of acceptance into the fellowship after about 1–2 weeks, and I don’t think I have to mention how happy I was to be included in the program and to be one of the 10 fellows selected from all over India.

Fellowship Program

The entire program lasted 8 weeks, beginning on June 8, 21 and ending on August 8, 21. The fellowship was divided into two phases. The first 4 weeks of the program were focused on exploring and learning about the Celo ecosystem and how one can one start building decentralized apps on the platform using blockchain. The next 4 weeks were focused on the building and development of the project that each fellow had to submit by the end of the fellowship under the mentorship of the assigned mentor.

📝 Week 1

In week 1, I learnt a lot about Blockchain, Smart Contracts, Dapps, creating an account on Celo, Using Celo to send transactions, How to run a Celo Node, How to deploy smart contracts using Truffle box and in general how is Celo different from Ethereum. Since I was a complete beginner, this was my first time working on Blockchain and Celo as a developer. In terms of learning, I got the basic gist of building on Celo and working on Smart Contracts and Blockchain in general. In terms of challenges, I faced some dependency issues while running the full node on Mainnet Network, and using the dappkit to get started with building mobile dapps on Celo because I was new to react-native and expo. Altogether week 1 helped me to understand the tech stack and the tools required to build on Celo and some basic concepts related in order to understand the working.

📝 Week 2

In week 2, we had to build a basic Dapp on Celo and present it to the rest of the Cohort by the end of the week. For this, I used the celo-dappkit with which I had worked in week 1 as the boilerplate. Since the deadline was a little tight, I decided to work on a simple use case that was to help users to store and update their health records using a smart contract to ensure immutability and security. For this, I created a smart contract with all the parameters like medications, allergies, diseases etc. and implemented functions to make changes to these parameters using the celo-dappkit in the front end. Since the deadline was extended, I tried to add one more feature. It was to add the buying of an insurance premium based on the user’s health record. I tried to implement that by adding another state variable called insurance followed by a function to calculate the insurance based on the age and other parameters provided by the user. Another function that I tried to add was a pay function to withdraw funds from the user’s account in case the user wants to buy the premium. Click here to view the project.

📝 Week 3

In Week 3, we had to research and go through any of the projects that were built using the Celo ecosystem as well as research for ideas for the project that we’ll be building further in the fellowship. For my research, I chose the project The Box which is a decentralized application built on Celo blockchain that makes it easy to collect money for any goal: collective gift, donations, fundraising, or even personal savings in cUSD or CELO. I chose this project because I wanted to learn more about smart contracts as well as how to use dappkit to make transactions using the Celo wallet. For me, it was a project that gave a complete overview of the functionalities that can be implemented using smart contracts and the Dappkit together. Below is the video of my presentation on the mentioned project.

📝 Week 4

In week 3, we had to give two project idea proposals that we were going to build using celo and work upon for the next 1 month in the project development phase. Upon researching and going through various ideas, I finally had two project ideas. The first one was Green Trip which was about offsetting the carbon footprints using celo cryptocurrency based on the transportation mode used. The second idea was HashImage which was a decentralized, censorship-resistant and image-sharing platform where users can earn in celo cryptocurrency by posting images in the form of art, photographs or even memes. After discussing it with mentors, I went forward with a second project that is HashImage as it was more extensive as a project and there were lots of features to explore in the given project.

📝 Week 5–8

After completion of 1 month, the learning phase was over and the next month was focused on project development based on the project proposal submitted. For my project, HashImage — I divided the work into weeks so that I have a structured approach. The first week was all about writing and developing the smart contract and getting to know about IPFS — which is a decentralized file storage system. The second and third weeks went into deploying the smart contract and the backend interaction with it using the celo extension wallet to transfer funds. I deployed the smart contract on the test network i.e. Alfajores network for demo purposes. The fourth and the last week was focused on the frontend development using react and wrapping up the project for submission.

You can check out the whole project over here:

HashImage

Conclusion

Overall the whole fellowship provided a great start for me to get into Blockchain development and understanding tools required to develop decentralized applications. These 8 weeks were truly extensive and exhaustive needless to say but the skills I acquired would remain with me for a lifetime. A big shoutout to all the other fellows, mentors, and devfolio community for helping me and guiding me throughout the program. Lastly, I want to thank Celo and Devfolio for giving me this opportunity.

We all are moving towards an era of Artificial Intelligence. Earlier when face recognition was something to be amazed at it is now easily implemented using existing libraries and frameworks. Machine learning is now embedded into our lives and it is thickening its grasp with time. Earlier it was a buzzword but now it is a reality that is making our lives easier and better. So let’s talk about some of the problems with Machine Learning. For those who don’t know about how machine learning works let me give a simple overview. It learns from the data provided as input to make decisions explicitly. In short, it makes its own set of rules based on the patterns it finds in the data. Now, what is this data that is being given as input? This data can be anything in any form like for cat and dog classifier it can a number of images of cats and dogs labeled accordingly. For house sales prediction, the data can be a spreadsheet with one of the columns as previous house prices and features such as the number of rooms, space, and other utilities as other columns. For house sales prediction, it predicts the house sales price by analyzing a pattern in the features of the house given. On the other hand for dog or cat classifiers, it extracts distinguishable features from the images of both the animals to label them. Deep Learning a part of machine learning enables machines to think like humans. Just like humans can easily identify a cat and dog, similarly, machines are trained to learn these from the images of cats and dogs.

Now if machines learn the ability to think like humans, do you think it would be effective if we humans didn’t have a stereotype. Stereotypes are oversimplified image or idea of a particular type of person or thing. For instance, Asians being better at math, African Americans being with greater athletic ability, Women supposed to be more emotionally sensitive than the men and the list goes on. Now the thing is stereotypes can be positive or negative depending on the way that particular community being stereotyped perceives it. But here is the thing, isn’t stereotypes based on the majority of people that exhibit the same characteristics or behavior. Asians are stereotyped as better at maths because a majority of Asians are good at maths. Similarly with other stereotypes too. So isn’t this majority a kind of pattern in the data. If with deep learning, machines are able to think and make decisions like humans, then do they also make decisions on the basis of stereotypes which is nothing but a pattern or distinguishing feature for the model.

Let’s start with the racial stereotyping. Predictive policing refers to the usage of mathematical and predictive analytics as well as machine learning algorithms to identify potential criminal activity. It can identify people as well as locations at greater proximity of a crime to happen. However, it received criticism for its racist behavior as it showed a trend to propagate racial profiling. American Civil Liberties Union’s (ACLU) Ezekiel Edwards claims that such analysis is more effective in forecasting police activities than in detecting crimes. In simpler words, it predicts how policemen would make decisions on identifying the potential criminals instead of identifying the crimes which are not wrong because it is trained like that. For instance, a policeman would be more suspicious of a person with ethnic race because of the stereotyping therefore the program also mimics the decisions of the policemen. According to an article in the Royal Statistical Society, it was stated that algorithms behaved exactly as they were expected to behave. They regenerated the patterns that were used to train them and even the best machine learning algorithms trained on police data would reproduce the patterns and unknown biases in police data.

Another example of racial AI is face detection and eye blink detection by cameras that highlight the smiling Asian faces as “Did someone blink” and in some instances do not recognize dark skin tone faces. If this doesn’t seem racist enough for you then how about when an African-American Couple was straight away identified as “gorillas” in their photos. You can search for these instances and you’ll definitely find a lot of them.

Why did the above scenarios and predictions take place? The answer is fairly simple. Unlike humans, AI doesn’t have any wrongful intention in doing so. It simply acts upon the data that is fed and how it is coded. However, due to the data being biased and inclined towards the majority, the AI-based technologies can sometimes display ignorance towards minority which leads to such racial scenarios and inappropriate generalizations.

Now coming back to the question “Does AI makes the decision on the basis of stereotypes?” and if it does “Would it be that effective if stereotypes didn’t exist?”. After reading the article what do you think and what is the solution for this?

Would AI and Machine Learning be that effective if stereotypes weren’t there? was originally published in Analytics Vidhya on Medium, where people are continuing the conversation by highlighting and responding to this story.

Earlier when no major storage device was present and there were only computers (hard disk) on which data could be stored, it was relatively easy for anyone to analyse or abstract information from this data. But with advancements in technology, there is an exponential increase in storage data ranging from removable sources such as USB sticks and SD cards to dispersed source such as Cloud Services. Therefore, big data can be described as the huge amount of structured or unstructured data stored in various storage devices usually in the size of zettabytes or petabytes that can be analysed to derive datasets providing patterns and interdependency of data.

Big data Challenges to Digital Forensics

Big data has been particularly challenging to Digital forensics where it has become extremely difficult to analyse data on three parameters of big data — Volume, Variety and Velocity. It is of no doubt how difficult it can be to analyse data of huge size that is being processed at a very high speed and having no idea from which device or source.

Furthermore, to find out complex interdependence and arrangement from this data would be even more difficult leading to an increase in cyber-crimes and a challenge for digital forensic investigators.

Proposed Solutions

Metadata

Data Volume problem can be solved by metadata which can be referred to as data on data. It can be divided as descriptive metadata and structural metadata.

For example if there is a database which stores details of a picture like the place where it was taken, at what time it was taken, what the picture was about? Then these details related to the picture is the descriptive metadata which would remain common for all pictures but the information that is filled in these details is the structure metadata like the place New Delhi, time 12 o clock and the picture of Qutub Minar which would be different for every picture.

Thus using descriptive metadata first and creating a structure metadata database from it can become a little less cumbersome to deal with huge amount of data.

How data science can be incorporated?

Data science can be incorporated in digital forensics to counter big data as it is nearly impossible for a human to analyse such huge amount of data and realise the complex patterns across the content. For instance, data can be collected from the previously solved cases, evidences and databases. This data can then be used to derive suitable algorithms set by the rules analysed on the given data. But in the initial test stages, the final decision should be made manually to avoid wrong decisions.

One example is of Predictive Policing. It refers to as the usage of mathematical and predictive analytics, and other analytical techniques in law enforcement to identify potential criminal activity. It was one of the best inventions as it was capable of stopping crime before it even occurred. But it had a drawback of discriminating against a group of people and often making wrong decisions.

Some of the tools from data science that can be made use of are MapReduce and Neural Networks. Apache Hadoop which is a collection of open source software utilities uses the MapReduce programming model to process big data. In this framework the large amount of data from the source is distributed into a number of computers called as nodes connected together so that the time taken to store data is reduced.

These nodes then replicate the data depending on the basis of application. Many top-notch companies such as Netflix uses Hadoop and Big data analytics helping it to save 1 billion dollar a year.

On the other hand, neural networks is an interconnected network of neurons that uses mathematical computation for decision making.

But in order to incorporate these changes the well defined principles of digital forensics need to be altered such as abandoning repeatability.

Moreover, in the workflow of the digital forensics several changes in the processes have to be made such as in the step of data analysis where data science would be required to evaluate algorithms and in reporting step, an accurate evaluation of all the tools and method used in the case has to be done and stored for future predictions.

Conclusion

It is estimated that by 2025 the world’s total data volume will hit 163 zettabytes. This can be taken as an advantage for business companies which would make use of big data analytics and data science to fuel their economy more than ever but will be equally challenging for digital forensics as the existing tools are not capable of handling such big data. Therefore with big data analytics and data science it is necessary for digital forensics to extend its already existing principles and research on new tools and technologies for investigative purposes.



Big Data, Digital Forensics and Data Science was originally published in Analytics Vidhya on Medium, where people are continuing the conversation by highlighting and responding to this story.