Machine Learning
& Artificial Intelligence
Next-gen software systems are not only functional, but they're also intelligent and perceptive —almost human-like in their abilities.

When most people talk about AI today, they're often referring to learning-based systems. These systems are mainly systems that become proficient (even human-like) at performing particular tasks — for example, self-driving cars and customer service bots. Learning-based systems "learn" by processing a vast amount of data (often labeled data, known as supervised learning). In contrast, at other times, systems learn by competing against themselves or other simulated agents.

Among all learning-based systems, neural networks have helped usher in a new wave of breakthroughs thanks to a technique called deep learning. Deep learning has recently been creating a lot of hype, confusion, excitement, and panic for companies looking to find a differentiating edge or niche.
Applied machine learning is a multi-stage process.

First, we need to have the data to learn from it, so gathering the data from various relevant sources (databases, knowledge graphs, etc.) is the first stage. Then, the data needs to be analyzed. The analysis includes understanding what the data can tell us and which features of the data are most applicable to the problem we are trying to solve.

Machine learning and deep learning go hand in hand. Deep learning has been touted for its extraordinary ability to automatically identify the relevant features. However, in practice, one will find that some hand-picked feature engineering and preprocessing goes a long way.

For example, suppose that we are trying to determine whether two faces in two different images or videos are the same. The naive approach is to train a deep neural network, e.g., via a triplet loss function. The Euclidean distance of images from the same identity is closer than the Euclidean distance of pictures from different identities.

However, the theoretical size (and thus required parameters) to automatically "learn" those features may result in an intractable network. This either results in the training becoming too expensive or the deployment nearly impossible (such as in resource-constrained environments where edge machine learning models may be desirable).
Being able to "crop" to only the face (i.e., preprocessing the images or video frames) can enable the final model to not only be more accurate but perhaps even more lightweight and faster, many times requiring fewer parameters to achieve higher accuracies. For example, in Kaggle competitions, the top solutions usually have a feature selection and preprocessing stage.
Suppose we are looking at 3D scans of an organ to determine the incidence of cancer. By preprocessing the data by identifying nodules in advance, this can enable the neural network to focus more wholly on areas of importance. This process significantly improves the final accuracy with the same number of parameters than otherwise. Then, it is time to train and evaluate a particular model.

This stage often involves research, usually scouring publications for similar or relevant solutions that can be reapplied for a new use case. If we were trying to detect whether specific abnormally shaped cells are precursors to cancer, we may look into the larger body of research that exists around melanoma, and reapply similar ML models as a starting point.

We may apply transfer learning from other datasets, seeing if some general learned features can be reapplied for the new use case. Throughout this entire process, we are continuously validating and evaluating the model, fine-tuning the hyper-parameters, analyzing the confusion matrix, and perhaps even performing grid searches.

Finally, the model needs to be deployed to the cloud. It can process requests and scale independently with the rest of the software application. The app then not only seamlessly integrates with the deployed model, but it also continuously collects data for future iterations and training of the model. We guide clients through every step of the multi-stage process.

How does a neural network learn?


One common type of question that we get is the form: "How does the neural network 'think?' How does the neural network arrive at its answer? Why does the neural network have that particular set of bias/weights?" This is a complicated question to answer. There is a lot of research being done to understand how a neural network "thinks" or "sees" or "hears," but first, it's essential to know how a neural network learns.

When you are a baby just learning to walk, you undergo a complex experimentation/trial and error process. Your mind is actively trying to minimize error — essentially, you are falling, getting up, falling, and learning. When we train neural networks, our goal is to either minimize or maximize an objective function, such as reducing a cost/error function. We are essentially "training" a neural network to respond in a particular manner when fed some specific input.

A neural network is mainly learning is how to map particular inputs to particular outputs. The neural network does this by "inferring" the answer for a specific sample of training. The neural network then performs gradient descent to update the entire system to "inch" towards (hopefully) the global minimum of the error function.

With enough training data, the network will get good at "generalizing" to new use cases. That is a basic introduction to neural networks in a nutshell.

AI is not just machine learning


Artificial intelligence includes more traditional approaches like rule-based expert systems (RBES) and behavior trees. These are often found in software systems that rely on domain-specific logic. For example, industrial automation is usually built on rule-based systems—many video games model complex NPC or enemy behaviors via behavior trees.

Machine learning is a subset of artificial intelligence, and Surya Tech has experience with both artificial intelligence and machine learning. Still, we especially value the potential that machine learning brings to our various clients. No matter which venture or product that we embark on with our partners, we ensure that data collection and data-driven design is always a central, guiding principle in the engineering of all of our products.

A vast number of businesses underestimate the value of their data


We find that a vast number of businesses underestimate the value of their data. In fact, many of our current clients viewed data as central to critical business decision-making, but that was where their fascination with data ended. We find that clients undervalue how valuable their data is. We work with clients across numerous industries to unlock the hidden potential of the treasure trove they are sitting on.

With startups and new companies, our experience tells us clients recognize the value of collecting data. Still, unsurprisingly, data isn't a central focus in the initial engineering of the product. This is a big mistake. Software and hardware / IoT systems need to prioritize data collection from the onset. Proper analytics and data-driven insights can help startups avoid costly mistakes and ensure that they are getting the best possible return on investment. This is essential at a period when every dollar counts for the new company.

At Surya Tech, we help and guide clients through the proper development of data-driven systems, collecting, aggregating, and reporting on data on a massive scale. From there, we train machine learning models to learn to automate numerous processes that once required human intervention. Whether you're in agriculture, medicine, the auto industry, retail, manufacturing, or even the government, AI has the potential to enhance how you work and engage with technology.

Our focus is on applied data science


At Surya Tech, our focus is on applied data science. This involves turning mainstream and bleeding-edge research into practical and novel solutions for our clients and their respective industries.

If clients wish to patent new ideas that we collaborate on, we will help you with that, too, from software to hardware!

We've helped clients apply AI across industries


We've helped clients automatically identify when crops are ready for harvest using machine learning and robotics. This process saved hundreds of man-hours per week from forgoing manual walk-by inspections. Additionally, we've helped clients in the staffing industry build AI systems to handle scheduling, recruiting, and payments. We've helped a client in the auto industry build a patented real-time digital auction system.

No matter your industry, we will guide you through the research and development process that comes with innovation. We will be your partner every step of the way, from conceptualization and ideation to prototype and design to MVP and beyond. We help you tackle the broader scope of your business as well, from the legal aspects (terms and conditions, privacy policies, patents, and trademarks) to marketing (integration with Google Analytics, Hotjar, Facebook / Snapchat pixels, Instagram, blogging/content marketing, and beyond).

Essentially, we help clients:

1. Identify novel research papers and publications that solve an analogous or similar problem.
2. Build innovative solutions, reapplying cutting-edge research to your industry's issues, from hardware / IoT, cloud, and AI.
3. Help you through any legal requirements (FDA, HIPAA, etc.)
We become your partners, not just engineers — we develop a passion for every project that we decide to tackle.
We work with
How we help
Identify novel research papers and publications that solve an analogous or similar problem.
Build innovative solutions, reapplying cutting-edge research to your industry's problems, from hardware / IoT, cloud, and AI.
Help you through any legal requirements (FDA, HIPAA, etc.)
We become your partners, not just engineers — we develop a passion for every project that we decide to tackle. Contact us today to start your next machine learning and artificial intelligence project.
Contact Us