The night sky has always captivated the minds of all those who have gazed upon it for centuries. For many a century, we have gazed at the millions of pinpricks of light that dot the dark tapestry and marvelled at the celestial dances they follow as they trace their perfect ellipses across the night sky. Space unites us as a species, as it brings the same unfiltered desire and yearning in all of us to explore its vastness and majesty.

Particularly close to our home planet, though, is our very own Moon. Since time immemorial, there have been legends and stories about the shadowy craters that dot the barren, monochrome desert. From Greek great chariots of war freckling the moon’s surface and the tapestry of an Asian rabbit cooking rice cakes to giants like the Apollo 11 mission, we as a species have always been fascinated with the lunar surface and how it came to be.

In the endeavour of exploring our crescent neighbour, space programmes have been at the forefront. With them, we explore the vast expanse of the cosmos, pushing the boundaries of human knowledge and expanding our understanding of the universe. Through a combination of scientific curiosity, technological innovation, and human ingenuity, space programmes have embarked on incredible journeys of discovery, unveiling the wonders of space. There are many famous Space programme like America’s NASA, India’s ISRO, and even private programmes like Musk’s SpaceX.

Recently, in the realm of space exploration, the remarkable Chandrayaan 3 mission has taken centre stage. Representing India’s ambitious foray into lunar exploration, Chandrayaan 3 stands as a testament to the nation’s unwavering commitment to scientific advancement and cutting-edge technology. Building upon the achievements of its predecessors, Chandrayaan 1 and Chandrayaan 2, this mission aims to unlock the mysteries of the moon’s unexplored terrain. Equipped with advanced instruments and state-of-the-art technology, Chandrayaan 3 seeks to delve deeper into the moon’s geology, mineralogy, and the presence of water ice deposits. Chandrayaan 3 represents the remarkable culmination of scientific expertise, engineering excellence, and technological innovation. It exemplifies India’s determination to explore the lunar frontier, broaden our understanding of the moon’s mysteries.

Anyone who even has the remotest interest in space programmes knows how challenging they are. Each mission is nothing short of a miraculous orchestra, a perfect balance of the sciences that propels metal tubes containing our hopes and dreams into the great beyond. With a little thought, one would realise that the amount of human interference on most missions past the pre-launch stage is practically zero.

One must wonder, though, how such magnificent machines perform such complex operations in outer space, where human interference is not possible? Most probes are sent out so far that it takes a few minutes and sometimes hours for the signals to eventually reach them due to the sheer distance! How do they manage to stay functional without us telling them what to do?

The answer to this intriguing question lies within the realm of computer science. It is through the ingenuity of computer scientists and the advanced technologies they develop that these magnificent machines can perform complex operations in the vastness of outer space, where human interference is not possible. At the heart of these space-faring machines are sophisticated computer systems specifically designed for the harsh conditions of space. These systems are meticulously engineered to withstand the rigours of the cosmic environment and ensure reliable operation over vast distances.

At this point, we wonder how exactly code would be used in such space programmes. Here are some of the primary areas where code is employed:

1. Mission Planning and System Design:

• Trajectory Calculations and Orbital Mechanics: Optimising Spacecraft Paths.
• Structural Design and Optimisation): Simulating Spacecraft Integrity.

2. Launch Operations and Countdown Sequencing:

• Launch Vehicle Control Systems: Real-time Control for Precision Liftoff.
• Telemetry and Tracking: Handling Data for Mission Monitoring.

3. Onboard Software and Guidance Systems:

• Command and Data Handling: Managing Spacecraft Tasks.
• Attitude Determination and Control: Ensuring Spacecraft Stability.

- Autonomous Operations: Real-time Decision-Making.

4. Ground Control and Mission Monitoring:

• Ground Station Software: Controlling the Spacecraft.
• Mission Planning and Analysis: Predicting Mission Outcomes.

- Anomaly Detection: Rapidly Identifying Issues.

5. Data Analysis and Scientific Exploration:

• Data Processing and Visualisation: Extracting Insights from Data.
• Scientific Instrumentation: Controlling Scientific Instruments.

An interesting factor to consider in such a mission is the harsh environment that space is. In the depths of space, unyielding background radiation poses a constant threat to the integrity of electronic systems. Cosmic rays are energetic particles that zip through space. They can penetrate sensitive electronic components and induce what is known as a “bit flip.” Essentially, a bit flip occurs when a single bit of data within a computer memory or storage unit is altered from its original state, potentially leading to incorrect calculations or erratic behaviour. The engineers tasked with the Apollo missions also had to face this problem. In this regard, the engineers came up with a brilliant idea.

Recognising the peril, the engineers behind the Apollo missions understood the need to fortify their systems against such cosmic interferences. They employed a strategic approach: redundancy. Rather than relying on a solitary computer, they integrated four computers. Yes! four computers, into the Apollo Guidance Computer (AGC) system, which played a vital role in navigation, guidance, and control.

This redundancy of four computers served as an ingenious solution to counter the effects of bit flips and other cosmic radiation. The computers operated in parallel, constantly cross-checking and verifying their calculations. By comparing the outputs, the computers could detect any discrepancies caused by bit flips or other radiation-induced errors. If a disagreement occurred, the system could correct itself, ensuring accurate data and reliable decision-making.

We must now talk about the interesting considerations that make modern computers successful. One key aspect is the autonomy built into these systems. Given the communication delays between Earth and the distant probes, it is crucial for spacecraft to make decisions independently. Computer scientists develop intelligent algorithms and decision-making frameworks that allow these machines to adapt to changing circumstances and respond to unexpected events.

Another vital element is fault tolerance. Spacecraft computers are designed to be resilient in the face of potential failures. Redundancy, where multiple systems work in parallel, is incorporated to ensure that if one component malfunctions, others can seamlessly take over.

Furthermore, the software running on these computers is meticulously crafted to be robust and efficient. Computer scientists optimise the code to minimise resource usage and ensure efficient execution of tasks. They conduct extensive testing and simulations to identify and eliminate potential software bugs that could jeopardise the mission’s success.

Additionally, the planning and scheduling of operations are carefully managed by computer scientists. They develop sophisticated algorithms that optimise the utilisation of limited resources such as power, data storage, and communication bandwidth. These algorithms allow the spacecraft to prioritise and execute tasks effectively.

Another interesting thing we must consider is the environment our code is being deployed to. When it comes to writing code to be deployed into space, there are a bunch of wacky rules that the developers must follow.

1. Strict Compliance: Space software development follows strict compliance with industry standards and guidelines. These standards provide a framework for designing, coding, testing, and verifying software to meet the highest levels of safety and reliability.

2. Minimalism: The code is kept minimalistic, with a focus on simplicity and clarity. Unnecessary complexity is avoided to minimise the chances of bugs, making it easier to analyse and verify the code’s correctness.

3. Robust Error Handling: Space software is designed to handle errors effectively. Developers employ robust error handling mechanisms to detect, report, and recover from errors, ensuring that the software can gracefully handle unexpected situations.

4. Defensive Programming: Defensive programming techniques are applied to anticipate and prevent potential failures. This includes input validation, boundary checking, and error checking at critical points in the code to prevent software failures caused by unexpected data or conditions.

5. Static Analysis and Formal Methods: Static analysis tools and formal methods are employed to analyse the code for potential issues, such as memory leaks, race conditions, and undefined behaviour. These tools help identify potential bugs and security vulnerabilities, enabling developers to address them early in the development process.

This article is just a brief foray into the wild world of space-proof code. Actual engineers use a variety of languages like C, Python, Fortran, Ada, MATLAB, Java, and Assembly for various purposes.

Coding is the backbone of launching space programmes, driving the success and efficiency of space exploration missions. From the initial stages of mission planning and system design to launch operations, onboard guidance systems, ground control, and data analysis, coding enables precise calculations, automation, and effective decision-making. As technology continues to advance, the role of coding in space programmes will become increasingly vital, empowering us to unlock the mysteries of the cosmos and push the boundaries of human exploration.

Celestial Code was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

Google introduced Federated Learning (FL) in the year 2017. It is a specific category of machine learning wherein its models are trained using decentralized data available on devices like mobile phones, self-driving cars, etc. It allows us to do machine learning while keeping the data on-device. It is resilient and very much secure.

Federated Learning allows for smarter models, lower latency, and less power consumption, everything while ensuring privacy. And this approach has an important benefit, i.e. along with providing an update to the shared model, the improved model on our phone can also be used immediately, powering experiences personalized by the way we use our phone.

How does federated learning work?

The significant insight is to realize that the nodes, which are the sources of training data, are not only data storage devices, but also computers capable of training a model themselves. The federated solution takes advantage of this by training a model on each node.

The server first sends each node an instruction to train a model of a particular type, such as a linear model, a support vector machine (SVM), or, in the case of deep learning, a particular network architecture.

After receiving these instructions, each node trains the model on its subset of the training data. Many iterations of an algorithm would be required for the complete training of a model, such as gradient descent, but in federated learning, the nodes train their models for only a few iterations, which means that each node’s model is partially trained after following the server’s instruction.

The nodes then send their partially trained models back to the server. And, they do not send their training data back.

The server combines the partially trained models to form a federated model. The average of each coefficient is taken to combine these models, weighting by the amount of training data available on the corresponding node. This is known as federated averaging.

The combined or the averaged federated model is then transmitted back to the nodes, where it then replaces their local models and is used as the starting point for the next round of training. After many rounds, the federated model converges to a good global model. In each new, the nodes can acquire new training data. Some nodes may even drop out, and new ones may join.

How is traditional machine learning different from federated machine learning?

1)PRIVACY ISSUES:

Traditional machine learning is based on a centralized approach where the training data is aggregated on a single machine or a data-centre. The well-known big companies like Google, Facebook, Amazon, etc have been doing so for many years. They collect a large amount of data and store it in their data centers where the machine learning models are then trained. But unlike the traditional machine learning method federated learning uses a decentralized approach, which is not privacy-intrusive like the conventional way. In a centralized approach, one has to trade their privacy by sending their personal information stored in their mobile phones to the clouds which can be accessed by the companies owning them.

On the other hand, in federated learning, the user doesn’t have to share any kind of private data to the cloud, and it allows mobile phones located at various geographical locations to concertedly learn a machine learning model keeping all personal data on the device itself. Hence it is much safer and secure in comparison to the conventional method.

2) EXPENSIVE COMMUNICATION:

Federated networks are basically comprised of a large number of devices (e.g., smartphones, smart vehicles, etc), and communication in such networks can be very slower than local computation. It is also much more expensive than the classical data center environments.

3) SYSTEM HETEROGENEITY:

Storage and communicational capabilities of devices in federated networks may differ due to variations in network connection, power signals, hardware, etc. The network size and systems-related constraints on every device result in a small fraction of the devices being active at once.

What are the challenges in Federated learning?

Federated learning involves learning a single, global statistical model from data stored on potentially millions of remote devices, which is beyond the thinking capacity of humans. To be exact, the goal is typically to minimize the following objective function:

m is the total number of devices, Fk is the local objective function for the kth device, and pk specifies the relative impact of each device with pk≥0 and ∑mk=1pk=1.

The local objective function Fk is often defined as the empirical risk over local data. The relative impact of each device pk is user-defined, with two natural settings being pk=1/mor pk=nk/m, where n is the total number of samples over all devices. Even though this is a common federated learning objective, there do exist some other alternatives such as simultaneously learning distinct but related local models through multi-task learning where each and every device corresponds to a particular task.

The multi-task and meta-learning perspective enable personalized or device-specific modeling, which can be a natural approach to handle the statistical heterogeneity of the data.

What will enable the growth of Federated learning?

In the coming years, model building along with computation on the edge, based on Federated Learning and secured with Homomorphic Encryption will definitely raise the peak of federated learning’s growth. As a large number of mobile phones equipped with AI chips and tremendous computing power will be available in the market in the coming years ahead, many ML models will be able to run simultaneously and locally on these devices. Distributing the heavy-duty analytics and computations over smartphones “on the edge”, as opposed to central computing facilities, will exponentially reduce time to develop data products such as hyper-personalized recommendation engines, e-commerce pricing engines, etc. Enterprises will go with a distributed machine learning model building framework for taking advantage of faster model deployment and which provides a quicker response to fast-changing consumer behavior, besides at a highly reduced cost.

For machine learning programmers, this shift provides a thrilling opportunity to customize AI. It also opens up new ways for adopting new tools and paving their way to deal with large-scale ML problems.

Though model development, training, and evaluation with no direct access to or labeling of raw data seems challenging at first, but in emerging markets such as our country (India), where hyper-personalization and highly contextual recommendation engine will be the key for app adoption and e-commerce advertisement, will play a huge role in the tech market, which indeed comes under federated learning.

Conclusion

Federated learning will create a future in which we work collectively to apply machine learning to some of the toughest problems that humanity faces, with each retaining control over our own data. It has the capability to solve the problems that the most regulated, competitive, and profitable industries face.

REFERENCES:

https://ai.googleblog.com/2017/04/federated-learning-collaborative.html

https://federated.withgoogle.com/

https://blog.ml.cmu.edu/2019/11/12/federated-learning-challenges-methods-and-future-directions/

FEDERATED LEARNING: THE FUTURE OF DISTRIBUTED MACHINE LEARNING was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

Ever wondered how companies increase retention and improve the User Interface experience for their customers? Or how they choose which content or picture to show on their main page? Or even which colour to make the button in their apps?

Well, in the current ever-increasing sphere of the internet, the answer (most of the time) is A/B testing!

What is it?

By definition, A/B testing is a way to compare two iterations of something that works best. In a simpler and more comprehensible way, it is basically an experiment in which users are randomly shown two or more versions of something (most of the time this is a website or an application), and statistical analysis is used to identify which variation works best for a certain conversion target group. Test results from this allow for better optimization of the website and permit for relevant data decision-makers to move market conversations from “we think” to “we know.” A really basic example for A/B testing would be, for example, to submit two randomized email versions to your client lists and determine which ones produce additional sales. From the feedback received from there, then, the winning version can only be sent later since we know that this was more engaging.

Another example would be if you check two ad copy versions to see which one is efficient at converting guests. Using the results gathered from it, you can know how & where to focus more time on getting the best returns. As you can see, the possibilities and uses of it is almost limitless. As long as you have a controlled & truly random sample base and are not performing more than one test at a time, A/B Testing is definitely the way forward!

History & origin

If we look back at its past, it seems apparent that A/B testing’s root can be traced back to James Lind’s ‘A Scurvy Treatise’ from 1753. It was found that in the 18th century, scurvy was the principal cause of maritime disease and death in the Royal Navy. In order to find the necessary antidote, Lind separately isolated six scurvied seamen and gave each a potential antidote remedy that various medical authorities had believed could possibly cure scurvy. Five pairs of seamen were prescribed vinegar, mustard and garlic purges, and other potential remedies. These seamen remained scurvied. However, for the remaining pair, Lind prescribed oranges and lemons & found that the pair quickly recovered, which proved that it was the best antidote. While the test seemed very rudimentary, it was in essence the first properly recorded example of A/B testing. This system is almost 100 years old and one of the simplest forms of randomized controlled testing, though these days it is most commonly associated with websites and application UI Development. By measuring the impact that changes in your UI have on your metrics, you can ensure that every change produces positive results which can enhance your business and growth.

Current scenario and uses of A/B testing

In the present generation, many say that the popularity of A/B testing has grown, as companies have started to realize that the online environment is well-suited to help them answer important questions like what are people most likely to click on, what are they interested in buying, or how do they like to register on the website? A/B testing is now being used to evaluate everything from website designing to headlines to even product descriptions. In fact, another surprising fact is that most of these testing and experiments run without the subjects (which is: you!) even knowing that they are a part of it.

If we look at some successful cases of A/B testing experiments which have been recognized as a large success, we see that one interesting & striking example is that of WorkZone, which managed to increase its leads through its testimonials page by about 34%. WorkZone is a US-based software company that was looking to revamp and increase its brand reputation. In order to increase sales, the company had put up a customer review section next to the demo request form on the lead generation page. Soon, they realized that the customer review section was hindering users from concentrating on the main aspect of the page, further distracting visitors from filling it. They decided to try changing logos to black and white and see whether the change would help increase the number of requests & surprisingly, it worked! They saw a massive spike in customer traffic, and it is safe to say that they adopted the updated lead generation page after this.

Now if we look at another example, Netflix comes to our mind, which went even further with their experimentation. The company makes different images for each and every video and tests these images themselves with a small proportion of their user base. The one which gets the most responses is called the winner and is then used as imagery for all of Netflix’s members. This is a very simple trick, but is one that has had a massive impact on the user experience. Video viewing for titles with images selected from such experiments increased by as much as 30%!

Looking beyond just the tests and their benefits, one important point to mention is that A/B testing is one of the most important components of the overarching process of Conversion Rate Optimization (CRO) — which is the practice to increase the percentage of website or device users who perform a desired action. Using this, we can gather both qualitative and quantitative user insights to enhance our online presence and business. You can further use this collected data to understand user behaviour, engagement rate, pain points, and even satisfaction with website features, including new features, revamped page sections, etc.

In conclusion, we can see that A/B testing is a very useful and important metric in order to achieve more efficient and better data understanding, especially for designing and implementing a better User Interface.

A/B Testing was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

This article has been written by OpenAI’s GPT-3 and compiled by Manav Goel, ISA-VIT member.

GPT-3 as an Artist

“Generative Pre-trained Transformer 3” is a machine-learning algorithm developed by computer scientists at OpenAI.

It is an open-source project which can be used to convert an image to an image which would best match the original.

Rather than focusing on illustrating one specific input with one specific output, this software is capable of generating a variety of possible results, similar to the human imagination.

Therefore, the generator might draw a picture of a cat while the input was a dog. This converter can even take a text word and add an image of that word to the generated image.

It was originally intended for use in video games such as DOTA 2, where the AI system can review itself and modify itself accordingly.

As with any artificially intelligent system, the training process is based on data received and their suggestions are based on their strategy. This system is therefore capable of replicating similar images as the previous images it has created, so that the image will be progressively more accurate.

Therefore, when an artist trains this system to make bird images for example, this system will not just look up images of birds and copy their shapes and colors: its generation process will be based on its understanding of what features are typically in bird images, and how different features are combined to create birds.

GPT-3 is also an Expert Linguist

OpenAI has an application for translating text to a language that does not exist. In short, it can translate any given language into English. This is necessary when translating a text from one language to another and the words are not in the vocabulary of either languages. For example, if I was trying to type “I love you” in French, I would need this application to translate it for me. It takes just one sentence for the application to turn the French into English text with all relevant translations necessary.

The Generative Pre-trained Transformer — 3 is a changer of languages or so it seems according to The New York Times article about OpenAI’s software. This machine learning system can write with phenomenal accuracy and is able to generate sentences of any jargon, including Russian, Mandarin Chinese and even Italian with no added training on specific languages. With each new sentence that is inputted into the system, it demonstrates how dynamically it changes its capacity when adapting to each different lingual structure and syntax. This could be a game changer for those who are interested in multilingualism or practicing a second or third language by being able to see what that text would look like before using this new technology-based translator installed on their computer or even tablet device of choice

OpenAI’s Generative Pre-trained Transformer — 3 AI software can translate from one text that does not exist and convert it into very accurate English versions within seconds using machine learning technology. Although there are many potential uses for this form of AI, there is much to learn and it may take some time before we see it in practical applications.

GPT-3 is the Future of AI

In recent years, machine learning has been used to build neural networks that can learn from data. These networks are called artificial neural networks (ANNs). ANNs have many applications in the fields of computer science, statistics, and cognitive science, but they are most commonly used in machine learning.

In the last few decades, ANNs have been used in a wide range of applications, such as handwriting recognition (e.g., Lenovo’s ThinkPad series), speech recognition (e.g., Apple’s Siri), natural language processing (e.g., Google’s search engine), and machine translation (e.g., Google Translate). ANNs have also been used to play games like chess and Go at a level that is competitive with or superior to the best human players, to recognize faces and identify human emotions, and even to predict the stock market.

The main difference between GPT-3 and current AI is that GPT-3 is what is known as “general purpose”. This means that it can solve any problem given enough time and data. Current AI cannot do this because it is only focused on one task or problem at a time. In addition, current AI requires many years of research before it can be put into use whereas GPT-3 only needs a few months of research before it can be put into use.

(All the text in italics is written by a human and everything else by the GPT-3.)

GPT-3 … a Polymath? was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

This product is essentially a mélange of an HRM (Heart Rate Monitor) and an ICM (Insertable Cardiac Monitor). In layman terms, a monitoring device such as this displays heart rate in real-time and monitors heart rhythm over a longer length of time than is provided by other tests or monitors for later study. Now, where is this device placed? On your wrist? Wrong! Such a device/ICM is placed just under the skin during a minimally invasive procedure in the chest above the heart.

Having an irregular heartbeat (arrhythmia) is more common than you might think. There is a 4x more chance that an irregular heartbeat can put you at risk of having a stroke. However, an irregular heartbeat can be easy to treat and monitor once diagnosed, thanks to the ample monitoring devices available. Now, wearables can track cardiac arrhythmias but they have their own limitations when it comes to making a ‘clinical diagnosis’. For example, many require recharging at night, which can cause the device to miss irregular heartbeats occurring sporadically and infrequently. Others require some level of interaction by the person to record symptoms, which may not be top of mind when feeling light-headed, dizzy, or experiencing a rapid heart rate. Even though the rise in heart rate monitoring has increased awareness about Atrial fibrillation and arrhythmia, most of these portable devices DO NOT lead to an improved diagnosis and sometimes fail to alert their respective care team to a potentially dangerous heart event.

So, unlike other monitoring devices, Confirm Rx ICM solves a plethora of such disadvantages and provides continuous 24/7 monitoring. The new detection technology in Abbott’s product reduces false detection by 95% and avoids alerting physicians of irregular heartbeats that may not actually signal a cardiac arrhythmia. The next-generation version of Abbott’s Confirm Rx ICM incorporates new SharpSense™ technology and Bluetooth® wireless technology, allowing patients to connect using their own mobile devices. Patients can use the “myMerlin™” mobile app to communicate with the Heart Monitor. One can also capture one’s own symptoms, using the app, when something doesn’t feel right.

With the introduction of this smartphone connectivity along with an easy-to-use app; Remote Monitoring has never been so engaging. This brings forth patient compliance with less disruption to daily lives. Earlier, physicians didn’t have favorable options when confronted with this dilemma. They could have the patient monitor symptoms on their own with a limited wearable device, such as a traditional EKG monitor, which can be a burden for the patient to wear day in and day out and time-consuming for the physician to obtain and interpret results, in part due to a lack of smartphone connectivity.

Micah Meulmester, Sr. Manager of Device Ops in Abbott’s Cardiac Rhythm Management (CRM) business, turned the engineers’ designs into reality by making the implanted device only 3.1 millimeters thick. “Size is an important thing for our patients because they don’t really want to know that it’s there and get tensed,” Meulmester says. “Because of its discreet size and smartphone connectivity, it’s something that patients don’t have to act on or worry about. The challenge for our engineers is how do you make a cost-effective product while also pushing the boundaries of technology?” It was an important innovation to understand packaging and manufacturing the technology in such a small device while also protecting the electronics from the body’s fluids and harsh internal environment.

Alex Soriano, Dir. mechanical development and material science in Abbott’s CRM business always mentions the astounding amount of unmet clinical needs at all points of time. To improve patient’s lives through engineering principles but also working smart, both Meulmester and Soriano came prepared for the challenges that the development of this next-gen device presented.

Since the first implantable pacemaker was developed in 1958, millions have benefited from pacemaker therapy. The remarkable story of the first cardiac pacemaker will be embedded in history as Abbott continues its commitment to revolutionizing arrhythmia detection and management with their state-of-the-art Confirm Rx™ ICM.

Abbott’s Confirm Rx ICM revolutionizing lives was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

1. Introduction

Car accidents or on-road collisions are something we witness almost daily on the news. The vehicle count on roads today is much larger than it used to be 10 years ago and as such, they are prone to more accidents.

The predictive analysis performed here aims towards analyzing the “Severity” of the accident/collision based on road conditions, lighting conditions, area of collision, the number of people involved, and many more factors like the aforementioned ones. Knowing the severity of any such collision beforehand will lead to prevention and prompt action.

2. Data involved

All the collision data used in this analysis is taken from ArcGIS, which was provided by the Seattle Police Department and recorded by traffic records. The data provided is that of collisions which took place in the city of Seattle, from the year 2004 till present.

Mentioned below is the list of features that were available in the raw data:

Feature List-

There are in total 38 data columns in the dataset including the 3 target related columns. We will keep various aspects in mind

while deciding the importance of a particular column or the transformation it may need before we feed it to the model.

Some of the given data columns are features related to or identifying a single particular accident, and thus may not be very useful for our predictive analysis. These features include:

SDOTCOLNUM, Coordinates, LOCATION, INCDTTM, INCDATE, REPORTNO, COLDETKEY, INCKEY, OBJECTID.

There are some description columns for a given code. Columns ST_COLDESC, SDOT_COLDESC, and EXCEPTRSNDESC are description columns for code which is already specified in the given dataset.

There are also data columns that have missing data in abundance. Column EXCEPTRSNCODE, EXCEPTRSNDESC, PEDROWNOTGRNT, SPEEDING, INATTENTIONIND, and INTKEY have more than 50% of data missing. Although few of these columns can be very crucial indicators of collision severity, it would be misguiding to use it with so many missing rows and very difficult to fill in these categorical values.

Columns mentioned in all the three categories above will not be used in the model that we are going to build. Most of the columns that remain are categorical and will require one-hot and label encoding before we can use them as a feature for our model.

3. Methodology

3.1 Exploratory Data Analysis

The first part of the process will be to explore the data and understand how a particular data column is distributed.

Most of our data columns are categorical and we need to know them to gauge the severity of the accident.

Frequency of Property Damage, Only Collision and Injury Collision with respect to collision type feature

Class distribution of ‘Matched’ and ‘Unmatched’ categories of the status variable

There are low cardinality categorical variables with 6–7 categories, moderate cardinality categorical variables with 40–70 categories, and very high cardinality categorical variables with 1500+ categories.

3.2 Feature Engineering

Mostly all the variables (except the features variable which defines the number of people, vehicles, etc.) are nominal features; i.e., features where the categories are only labelled without any order of precedence. The preferred encoding for these categories is One-Hot encoding. However, One-Hot encoding will generate around 1500 data columns for just one high cardinality categorical variable, which will be very expensive to work with.

We can get over this hurdle by using feature hashing. Feature hashing is an encoding technique which is used to encode high cardinality feature by hashing them. By this, we can pull down the number of encoded data columns to 32–64 even for variables with >1500 categories.

Distribution of all missing data in the training set was found to be:

Distribution of all missing data in the training set

As the class proportion is not getting much affected by dropping these data rows, we will proceed to do so.

After the process of feature hashing and one-hot encoding, we obtain a total of 208 feature columns. We are using Random Forest to get the feature importance, eliminating 40 least important features and correlation matrix to detect >90% correlations.

After removing the least important and highly correlated features we are left with 160 features to train the model with.

3.3 Modelling

As it was clear from the above analysis that we have had a skewed dataset. This resulted in a low recall of class 2 and as a result low F1 score.

To solve this problem, we used smote to oversample the rare class and generated the cross-validation score again. While doing oversampling we have to keep in mind that oversampling should be done on each iteration of cross-validation and not on the whole training set.

As a result, we observed that although the recall in class 2 and F1 increased a little bit, it decreased the accuracy too. Considering the increase in computational expense due to increased data, oversampling didn’t prove to be worth the effort in this case.

We used the XG Boost Classifier to start with and plotted the learning curve to see if the model is overfitting the training data.

We observed that converged training and validation errors were close to each other, which means that we can use high variance algorithms like Random Forest, XG Boost, and Support Vector Machine, and we can also use the high number of features that we are using.

As expected, we got the best performance from XG Boost Classifier. We will further try hyperparameter tuning to improve performance.

4. Results

For the final prediction, we have to preprocess the whole test data-set. While encoding the feature columns we made sure that the one-hot encodings are the same as the train set and the feature hasher transformer used should be fitted on train data.

Following are the Final result of the test data:

5. Discussion

Many more analyses and methodologies can be added to this project in the future. We haven’t used the coordinates. Those coordinates could result in some unforeseen clusters which could exponentially improve the study.

Further other encoding techniques can be used in place of feature hashing or feature hashing with different feature count can be used. The performance of these changes can be evaluated using cross-validation.

6. Conclusion

The results are satisfactory but expectations were much higher. A lot of improvement can be done on class 2 predictions. Overall a lot of improvement can be observed from the basic model.

7. GitHub links :

Refer to the following links for further understanding of the project:

https://github.com/AryaPatel1111/Data-Science-Capstone

https://github.com/AryaPatel1111/Data-Science-Capstone/blob/master/Final_Notebook.ipynb

Preventing Car Accidents using Data Science. was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

Artificial intelligence and machine learning are all the rage these days but have you ever heard about 3D printing? It is one of those new technologies that is currently spreading like wild-fire due to its effectiveness. It can create products without any human intervention just like a CNC machine, but better. Better in terms of the raw material used, achievable accuracy, reduction of defects, and human errors, the list goes on and on. There are a lot of things you can make with the help of a 3D printer, provided your product fits inside the permissible workspace. But today, we are going to discuss the jigs and fixtures using 3d printing tech.

You might have heard about the headlines that “BMW realized a cost cut of 57% by using 3d printed jigs and fixtures” or about Dixon Value, which made a whopping 97% reduction in manufacturing cost using 3d printed grips and jigs. But what are these jigs and fixtures? Jigs and fixtures in a broad sense are production tools used to hold the workpiece to produce duplicate or interchangeable parts. They differ in the sense that in a fixture is used to support or hold the workpiece; on the other hand, a jig is used to do the same in addition to guiding the tool to the workpiece.

Here the U-shaped body is a fixture, it will only support the workpiece. As there are no boring holes, so it will not be able to guide the boring bar.

This image is similar to the previous one but instead of a fixture, we have a jig here. It will hold the workpiece and in addition to that, it will also be able to guide the boring bar.

So now we have a basic idea of what jigs and fixtures are and how they are used in the industry. But why would anyone 3D print this structure when I can make it with traditional tools?

Time is Money:

Traditionally, a jig or fixture is made from aluminum or steel and they undergo a lot of machining process before being employed in the manufacturing line. The whole process can take up to 15–20 days using standard machining tools. This same jig or fixture can be made with the help of a printer in just 1 or 2 days with a high level of accuracy(Industrial FDM — ±0.2 mm, SLA — ±0.05 mm, and SLS — ±0.1 mm), and this is an enormous cost saving for the manufacturer in terms of the material selected and less production time.

Variety of materials:

The materials used in 3d printing are generally polymers, though If required you can also make metal products, but that’s a different story. This polymer exceeds many metals in terms of chemical resistance, heat resistance, UV light stability, shock resistance, and many more. This polymer also provides soft contact with the workpiece. Using a metallic fixture can cause marring or scratch to sensitive products.

A safe worker is a happy worker:

Jigs, fixtures, grips, and other tools require manual handling by workers, and creating lightweight parts that can be easily transported is one of the main aims of a manufacturer. As tools are designed ergonomically, which will mean a safer environment for the workers and will further contribute towards the overall production efficiency of the plant.

3d printed tool can be a lot better than the traditional tools which were clunky, expensive, required a lot of machining, and a crazy amount of time when compared to a 3d printed alternative. Industries are already making use of this technology by replacing traditional metal counterparts where ever possible.

Jigs and Fixtures using 3D printing was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

-By Suyash Gupta

This novel and exceedingly innovative technology is a prime example of modern technology blurring the lines between science fiction and reality. Innovations like these are truly a testament to mankind’s ceaseless pursuit of becoming the master of its own destiny and a testament to our undying desire for constant improvement.

What this Idea Aims to Accomplish — An Overview –

Neuralink, founded in 2016 by Elon Musk, at its core aims to create an effective interface between the cortical region of the brain and AI or a symbiotic relationship between the human brain and AI. This in turn means that it can pave the path to various breakthroughs such as being able to upload and download thoughts, allowing the user to control wireless devices through mere thoughts. This could prove especially useful to paralyzed people, allowing them to use such devices despite their physical limitations.

This technology of brain and computer interfacing has been done before, however, they have certain issues that hinder their viability as stiff wires and electronics within the brain can lead to inflammation within the tissues of the brain. Neuralink, on the other hand, implements flexible floppy wires which can be safely embedded within the brain for longer periods without sustaining damage to the soft tissues.

Foundations of the Technology –

Technology which allows direct interactions between human brains and machines are not an entirely new concept, however, Elon Musk’s neuralink builds upon this foundation of existing technology to provide a unique approach to brain human interfaces to make them more viable, less invasive and have a wider reach.

Some examples of previously implemented Brain-Computer Interfaces (BCIs) being put to use can be found in the early 2010s. For instance, in 2012 paralyzed patients were able to control a robotic arm connected to their brain. A way in which neuralink’s advanced can enhance such a function is by providing a sense of touch simulated in the robotic arm provided to the patient in question.

Implementation and what’s New –

The first stage in the implementation of Neuralink aims to reduce the invasive nature of current brain-machine interfaces. This step improves the manner in which the system is attached with the brain, as it uses a machine akin to a cross between a microscope and a sewing machine, using which it attaches various thin thread-like flexible conductors with a diameter about one-third that of hair to the brain’s neurons.

These conductors were implemented into the head of a rat and were allowed to feed data through a USB-C port attached to its head, and in doing so it provided more than 10 times the amount of data provided by the current best sensors in the market. Furthermore, the chip equipped with a USB-C port has been developed as a custom chip that is better suited to read, clean up, and amplify signals from the brain.

Plans for the Future –

The aforementioned method of using thread like conductors installed via a so-called “sewing machine”, in its present state has some limitations as holes must be drilled into the subject’s skull in order for the conductors to actually be placed. Keeping this in mind, future iterations of this machine will be developed to instead use lasers to penetrate the skull to allow the technology to be installed.

Additionally, the chip mentioned above is set to be modified in a manner so as to facilitate wireless data transmission via a product referred to as “N1 sensors”. The plan is to embed four of these sensors into the brain; three in motor regions and one in a somatosensor, which will be connected wirelessly to a device mounted behind the ear.

Above all this, however, the main plan for Neuralink is to read and amplified Neural spikes in the brain in a minimally invasive way paving a path to a successful merging and symbiosis with AI.

Obstacles faced by Neuralink –

So far, no tests have been carried out on human subjects owing to obvious ethical concerns. There are concerns that such a technology could compromise the free will of a person equipped with it, but, there are methods to subvert these issues, such as but providing protocols and two-way encryption between the brain and the machine.

It was planned for the technology to be tested on humans after refinement of the technology in late 2020; however, that may end up being delayed owing to the current unfortunate global circumstances. Besides, in order for such testing to be underway, the organization will have to gain approval from the FDA to proceed further.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

Neuralink, at the end of the day, is an excellent innovation that builds very well upon the foundation laid by formerly implemented technologies of its ilk. It could easily prove to be a huge leap in the progress of humanity and the augmentation of people, exploring never-before-seen facets of biotechnology.

It is without a doubt exciting to see this technology being augmented upon and for it to take a proper foothold in the markets of the future as it reaches the pinnacle of innovation.

References:-

https://techcrunch.com/2019/07/16/elon-musks-neuralink-looks-to-begin-outfitting-human-brains-with-faster-input-and-output-starting-next-year/

https://www.biorxiv.org/content/10.1101/703801v4

https://www.theverge.com/2019/7/16/20697123/elon-musk-neuralink-brain-reading-thread-robot

https://towardsdatascience.com/elon-musks-neuralink-everything-you-need-to-know-a19c35708f9a

https://www.businessinsider.in/tech/news/inside-the-science-behind-elon-musks-crazy-plan-to-put-chips-in-peoples-brains-and-create-human-ai-hybrids/articleshow/71463463.cms

Neuralink — A link between man and machine: was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

-By Suyash Gupta

Planty Cube is a revolutionary step forward for the novel and innovative idea that is vertical farming and can even be considered the next evolutionary step for farming systems in general. This ingenious application of embedded systems is the epitome of the pervasiveness of IoT applications in modern industries. This blog aims to explore this idea of vertical farming, the necessity for its implementation, and how it can be enhanced using IoT highlighted by this example system — Planty Cube.

What is vertical farming? — And Why do we need it? –

Vertical farming is a new method of cultivating crops born as most innovations are — out of necessity. As the population of humans grows further every day, expected to grow and eventually plateau at the staggeringly large figure of 9 Billion, A major challenge this presents us with is the sustainable management of resources, especially crops and food. Conventional farming methods are not very efficient as these methods require vast amounts of arable land which is running low at an ever-increasing rate, and also huge volumes of water are used up for watering plants out of which only a small percent of it is actually necessary.

Vertical farming aims to solve these issues by adding an extra dimension to the space in which crops can be grown and also use efficient methods in terms of resources used and space taken to grow crops.

What it would take to feasibly implement a vertical farming system –

Vertical farming would be implemented in stacks within indoor settings such as warehoused and skyscrapers and would be built in appearance, not unlike an indoor form one could make in Minecraft.

However, there are some critical factors that play into making such a system feasible such as the layout — which most often involves horizontal planes stacked vertically producing an increased yield. The most crucial element of these systems is arguably the lighting required by the plants in order to grow, and since there have been many advancements in led technologies, these can be implemented efficiently to provide the necessary radiations in the form of wavelengths in the blue and red range for the effective growth of crops. And finally, the medium in which crops will be grown will most likely involve hydroponics in which a plant will be submerged in a nutrient-rich water solution which could be circulated and reused.

How IoT can enhance this already inventive idea –

The implementation of IoT will allow for vertical farms to be more modular, having settings for different breeds of plants to be grown in such systems, and in addition to this, the computerization of such systems will allow for the monitoring of the health of the crops being grown.

As such, this data can be collected within a database and be used to formulate the appropriate settings required for the plant to thrive. The system would be able to control the lighting provided to the crops, the irrigation, humidity, and temperature to optimally suit the plant in question. This system will allow for the efficient growth of crops and would be able to decide the right growing conditions for the crops in a smart manner, serving as a fully automated sustainable farming system that is easy to maintain.

Challenges faced by automated Vertical Farming systems –

While, it is not difficult to understand the numerous advantages that such vertical farming systems can have, there are also a few apprehensible issues in these systems which must be addressed. Integrating something like this with IoT will lead to us having greater dependence on networks and technology which could lead to large-scale issues when systems like this fail or experience bugs of some sort. In addition to this, the added complexities involved would result in a greater need for maintenance.

How far this Idea has come, and what we can expect in the future–

Despite the aforementioned issues mentioned these systems’ benefits far outweigh the negatives and as such this system has already seen a fair amount of success. For instance, AeroFarms has repurposed a steel mill to serve the purpose of vertical farming as its largest iteration in the world and now harvest up to two million pounds of crops each year.

This is a prime example of how this idea can be brought into the mainstream and in the future, we might be able to see even more companies adopt this style of producing crops, perhaps with more enhanced IoT systems to assist the farm. And in the near future, it may not be too absurd to see entire skyscrapers dedicated to this ambitious and ingenious purpose.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

In conclusion, implementing vertical farming methods and even having such systems replace traditional farming, in the long run, would allow us to save more land and resources in the process of producing crops. This, in turn, would allow us to have more healthy yields and be able to provide greater accessibility to food for most of the world’s population. Further accentuating this idea and enhancing vertical farming systems are IoT based solutions such as Planty Cube at the forefront of this revolution, allowing the automation of these systems, making the management of such systems easier and helping the operation of such farms massively.

Such an Idea will undoubtedly bring promising and bountiful results in the future and will completely reshape our idea of farming and producing food.

References:-

https://www.youtube.com/watch?v=IBleQycVanU&list=PLBTrWbLcO_tkuW44Zr4XSwDD9Ekqm6n4V&index=4&t=4s

https://aip.scitation.org/doi/pdf/10.1063/1.5002039

http://www.mintcontrols.com/vertical-farming-iot/

https://www.designboom.com/technology/the-planty-cube-vertical-farming-system-ces-01-14-2020/

https://www.researchgate.net/publication/319867877_Vertical_farming_monitoring_system_using_the_internet_of_things_IoT

https://edgy.app/planty-cube-offers-smart-farming-alternative

https://www.theburnin.com/technology/planty-cube-vertical-farming-system-revolutionize-agriculture-2020-1/

https://thespoon.tech/ces-2020-the-planty-cube-aims-to-make-vertical-farming-more-modular-and-automated/

Vertical Farming — the next big conquest for IoT: was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.

The internet is no longer a web that we connect to. Instead, it’s a computerized, networked, and interconnected world that we live in. This is the future.

-Bruce Schneier
(security technologist and author)

In recent times, the theory and actual practice of cloud computing has gained momentum. Computer systems, remote devices and appliances now come with this technology in-built. Explication of edge computing is the practice of processing data near the edge of your network, where the data is being generated, instead of in a centralised data-processing warehouse.

There has been speculation on the usefulness and user-friendliness of edge computing and its features. But the recent development in the fields of mobile computing and Internet of Things (IoT) technologies has made the world see the massive scope as well as its effectiveness.

Edge computing refers to the processing of data of the Internet of Things closer to where it is created. In this way, data is processed at the edge of the network, by performing analytics and knowledge generation at or near the source of data.

Edge computing works by pushing data, applications and computing power away from the centralized network to its extremes, enabling fragments of information to lie scattered across distributed networks of the server. Earlier available to only large-scale organizations, it’s now available to small and medium organizations because of the cost reductions in large scale implementations.

For instance, consider a building fortified with high-definition IoT closed-circuit television cameras. These cameras simply output raw video signal and continuously stream that signal to a cloud server. The cloud server at its end puts all video outputs into a motion detection application and only those clippings are stored in which some sort of activity is observed. Now, streaming such high definition signals uses significant bandwidth and also levy sizeable load on the cloud server that has to process the video footage from all the cameras simultaneously.

Now, putting edge computing and its ancillary technologies in place of the given scenario. If each computer has its own small-sized internal computer to run the motion detection on its own and send only the clippings where some sort of activity is recognised. This reduced usage of network bandwidth many folds. Also, this relieves the cloud server from a heavy load of processing videos, as a result, it can store relevant clippings of more number of cameras conveniently. Bringing the majority of the processing to the source of input is what is edge computing.

Another advantage of moving the processing to the edge of networks is that it reduces latency. Consider the time delay which comes into play when a remote device has to send signals to a far off cloud server. In comparison to that, if the device has its own internal processing unit, a lot of time and therefore money can be saved.

When one has to deal with a lot of data, you leverage IoT in such end-to-end ways or even in specific highly sensor-intensive and thus data-intensive environments whereby data is generated at the edge which by definition happens in IoT as your data sensing and gathering devices are at the edge. You inevitably encounter challenges on levels such as bandwidth, network latency, speed overall and so forth where primitive technologies play a role. In IoT applications with a mission-critical and/or remote component, the need for speed and for different approaches such as edge computing is even more important.

You need the aggregated and analyzed data, in the shape of actionable intelligence, enabling you to take actions and decisions, fast, whether these decisions are human or not. So, you don’t need all that data to store it and analyze it in the cloud but you only want that bit of data travelling across your networks.

You can imagine hundreds of scenarios where speed and fast data is key, from asset management, critical power issues, process optimization, predictive analytics to the real-time needs of supply chain management in a hyper-connected world, the list is endless.

You can also imagine that the more your building, business ecosystem and whatnot thrive on fast data and real-time holistic management in any broader context, the more valuable that data can become when properly leveraged and rapidly analyzed. We do live in times where having the right insights fast enough can have enormous consequences.

Speed of data and analysis is essential in many industrial IoT applications but is also a key element of industrial transformation and all the other areas where we move towards autonomous and semi-autonomous decisions made by systems, actuators and various controls.

That degree of autonomy is even at the very core of many of the desired outcomes.

References

1. www.networkworld.com
2. www.hpe.com
3. www.enlightened-digital.com
4. www.cloudflare.com

Originally published at https://blog.isavit.club.

Edge computing was originally published in ISA-VIT on Medium, where people are continuing the conversation by highlighting and responding to this story.