Thank you for attending this session and, let me first give you a quick introduction about myself and what I do. so I work at a tech company in Seattle, and I have more than eight years of experience in, data engineering, analytics, machine learning, and cybersecurity. I graduated from Northeastern University in, and I have been working in this field and I, apart from work, I like to travel, I like to also try to learn new AI models and try to fine tune those models. So today we are going to cover presentation on AI frontiers, deep learnings, transformative impact. We have seen a lot of deep learnings advancements in reshaping industries using reinforcement learning, generative models, and these are being used in almost all the sectors, including cybersecurity. So let's dive deep into it. so this slide provides you about reinforcement learning. So what do you mean by reinforcement learning? So reinforcement is something you try to reinforce something and try to make it more powerful. So that's exactly the concept of reinforcement learning. So it is like a reward based learning and it helps the AI based system to overcome the errors by using trial and error technique. And most of the algorithms in this space use pattern recognition and using deep learning and strategic decision making and neural networks and all these kinds of models. So they are based on reinforcement layer based learning, where you provide the rewards and then using trial and error method. It tries to generate a pattern and improves the model as compared to the supervised and unsupervised models. And reinforcement learning is very powerful and every year we are seeing a lot of new advancements using reinforcement learning. Next slide over here is very important. It goes through the transformer based model. So transformer based model, why it is so powerful and why it is being used. So let's dive deep into a transformer model. So transformer model is basically a neural which kind of learns the context, by tracking the relationships in sequential data and words. And it helps to overcome a lot of challenges and solves a lot of problems. Problems, and it is a very fast working model. It can be used in time series analytics. It can be used in recognition. It can also be used in kind of image and image detection, as well as words, predicting the words and translating a sentence and a lot of other use cases. There are, so in this slide, let's go deep and understand the architecture of a transformer. transformer is basically encoder and decoder algorithm. And, how it works is, it has, it has, these kind of attention function where first it encodes the word. Like for example, we look into a particular example where we have to. Translate a sentence into another language. So in this case, what will happen is we'll provide the inputs in input embedding. And from here in the encoder, it will go into the multi head attention. So attention function, it is a very important function over here, and it tries to take all the components, let's say in the sentence we have eight to 10 words, so it will create these kind of eight to 10 vectors and these kind of vectors they are quite, it will optimize and make these vector and then it will go into this decoder function where it has feed forward multi head attention, and it also has masks, multi head attention, and it will then try to translate the word and then find out in this use case, it will try to find out in the language which we want to translate. So it will try to generate the words in that particular language. And then finally it'll, try to optimize and, generate the output words. So this kind of, algorithm, why it is being used and very popular. because this transformer based, ai, model, this is very faster and powerful and it works, parallelly. So other models we have seen, they do not work. as parallel as this model and, transformer based, machine learning models because they work way in parallel, it can utilize the GPU like Nvidia's GPU and all kinds of GPUs in a very efficient way. So the GPU is utilized in an efficient way and it can work very faster and it has the ability to scale up and work very fast. and the data sets which are being used are scalable and quite scalable and you can get the output quite faster. Then any other, AI algorithm, CNN, RNN, or any other algorithm, which you can use. So another use case over here is it can predict the next word. So based on the words you write, the algorithm will be able to predict. Or what would be the next word? So it would be in the same situation. It'll use the encoder and, decoders attention function, and it'll try to edit the next word. Now we come to time GPT. So what is time? GPT? So time, GPT is, GPT based model. It's a transformer based time series analysis model, which was developed by nla. So this. Small startup. It has created a very good AI model and it can be implemented by a very small piece of code. So it's a production ready generative pre trained transformer for time series. So this model can be utilized to you. create time series analysis and all kinds of things. So you can look into all the other time series analysis and then try to use this model and it will provide you with more accuracy and more accurate results and it is much faster scalable and it can help you to optimize the time series and it just uses these kind of small lines of code where you import the Nixla client, use the API key over here, then you utilize in the Nixla client, you try to use that and you load the data from, Python. Any large data from URL, or a CSV, and then use that data frame and use the next Nixla client to forecast and then use the forecast function to develop the time series. So implementation is much faster as compared to any other time series analysis. And it is very efficient to use in detail finance, IOT based applications in tech. or cyber security as well. So now we come to another slide. This slide provides us with the information about convolutional neural network. What do you mean by convolutional neural network? So basically, this kind of neural network very popular and it is being this kind of neural network is being used for image related tasks. It can also be used for autonomous driving where it can analyze and detect various obstacles on the road. It can find out if there are any people around the road and detect various things on the road. So basically it is optimized for image related tasks and it can also be used for video games and such kind of applications and this is, these are the key components of convolutional neural network. Is the convolutional network, then the third one is the fully connected, layer over here. And, we have the activation function where it improves and creates the output. So first we come to the pooling layer where it Reduces the dimensionality of the feature by summarizing the re regions and by using techniques like max pooling, and this helps us to retain the important information and also reduce the complexity. Then we have the. Pooling, the fully connected layer, the fully connected layer is where the high level features within the function are used for classification, regression, and various other tasks. So in the fully connected layer, we use high level features and it can be used to optimize and create more better outcomes. Then we have the activation function. There are a lot of activation functions which are used to improvise and introduce some non linearity into the network and then it also improves the neuron's capability and those kind of things. So convolutional neural network is also quite extensively used, in, neural networks. Many of the use cases like self driving cars and image related tasks and for photography image and such kind of optimizations convolutional neural network can be used. We also have recurrent neural network which is another kind of a neural network as compared to convolutional neural network which I don't know. Is mostly used for applications related to anomaly detection and more kinds of patterns kind of detection, and mostly it is being used in cyber security. Okay, so now we come to this slide over here, which provides us information about GA and BAs, so generative adversarial networks. these are generative kind of, Algorithms and VAs are variational auto encoders and, they have different use cases right now and they are very popular in anomaly detection. So VAs are very important in anomaly detection. It can, anomalies, within cyber security. So the good use case of VA is are in cyber security where it can, detect various kinds of threats, improve the accuracy of your model, existing model, and, analyze the patterns. in cybersecurity, but mostly you need to first use your, current tools. You have SIM tools and various other tools in cybersecurity, and then try to use VAs and you can also use R Ns for finding the anomalies. Then we have photo realistic images, so advanced styles for image generations and all kinds of things is AI algorithms can be used for photo generating photorealistic image. Now we come to the next slide. So this slide provides information about the AI powered thread detection. and using graph neural networks. So why do we need AI based threat detection? Because currently the threat detection capabilities using cyber security tools are a little bit limited and as more and more threat actors are evolving, there is a constant scenario where they can use AI based tools to plan an attack, DDoS attack and such kind of things. In this scenario, if we use AI based threat detection, it can help us to mitigate such kind of threats and provide us more patterns and with results with improved accuracy and also reduce the false positives. But in this case, we have to train the model. In cybersecurity and compare the results with the existing tools we have Splunk or Exabeam and such kind of tools and try to find the threats and compare the threats with the AI based algorithm and then try to find out which would be the best algorithm to use. Then we have graph neural networks, which are, shaping and, detecting more of security based monitoring, finding out the patterns and improving the detection capabilities and conventional methods. So some of the important tools maybe in next 5 to 10 years, we might see a lot of, use cases of these tools and these tools being applied in security automation and improving the threat detection capabilities. So now the next slide provider provides us with the AI convergence and future direction. So we have what were here, multi modal AI, which is now seamlessly integrated with multiple forms of input, vision, speech, sensor data, creating more, Intuitive and comprehensive solutions, which are similar to human understanding. Then we have Age AI, which is more of algorithms working directly on local devices rather than the cloud. And Age AI, because it is working on local devices, it is very fast. And provides us with the good amount of, privacy and more kind of faster response and revolutionizing applications from autonomous vehicles to healthcare devices. But the, but now when we apply such kinds of things in our, use cases, we need to understand the ethical, considerations. You have to work with the compliance team within the company to make sure that the AI models are working well. They are not taking any customer data or changing anything within your environment. So be careful when using such kind of AI models in cyber security or such kind of other use cases you have and try to make sure that these kind of AI models are not having any impacts about. Privacy or, any changing any data governance framework. So before deploying, so please try to make sure the deployment of AI algorithms are done in a more ethical consideration. And now we come to AI's transformative impact on industries. AI will have a lot of transformative impact on the industries. It will be having autonomous systems and of self driving vehicles with 99 percent or maybe 95 or more percent accuracy. Then it will be used in drug discovery and it. will try to create, drugs at a more cheaper price as compared to currently what we have. And it will improve the pace in which the drug discovery is being made. So there will be cost benefit. There will be a lot of mass production of drugs. Then we have, Medicines like these kind of AI tools will generate a lot of data points for the patients, then take the inputs from the patients, from the doctors and plan the treatment. So AI based tools will provide a lot of personalized medicines. It will help the medical sector a lot along with pharmacy and autonomous systems. So all kinds of these kinds of algorithms like CNN, RNN, transformer based models, they will help to improve, various systems, along with the drug discovery and medicines. So next we come to, breakthroughs and practical implications. So deep learning algorithms, have. evolved and they will try to evolve, again and again, and they will evolve further in next five years. So right now we have transformer based models. Maybe in next few years, we might see more faster, models than this, and, it will help us to create more, Better outcomes and take, data from various diverse sources and try to create a better kind of, outcomes with accuracy. now we have, the AI advancements. probably in financial intelligence, then we have customer intelligence and, various, various other places. I can help us to reduce the down times and response times, and, it can also provide patterns, detecting patterns with accuracy and enabling data driven strategies. And, it will have also help us in recommendation and such kind of things. And we have to make sure that in all of these things we do, we have to keep in mind the ethics and governance. We need to make sure that the governance, data governance, and AI based ethical considerations are looked into so that they don't exploit anything or, they change anything or put they can take the customer data and utilize it and, have, bigger kind of, issues. working with compliance team, data governance, and ethics. ethical consideration also needs to be looked into. So when you look into, when you look into such kind of things, AI models will improve the accuracy and they will evolve. And maybe by 2030, you might see better outcomes using AI based algorithms, using quantum AI and such kind of things. that's it from my, side. thank you for attending this session. please do leave your feedback and also try to connect with me on LinkedIn. If you need to get more information, I can help you out.