Pradeepta Mishra

Head of AI at LTI


  • Can you tell me more about TensorFlow and whether it is a good option for speech recognition?

    for speech recognition you should use Keras APIs it's much better than TF.

  • When and How did you start your career in Machine Learning?

    My ML journey started from Prof. Andrew Ng’s course on Machine Learning from Coursera, I am from the first batch (2012-13) when the course was delivered using Octave and Matlab as a programming language

  • What motivated you towards Machine Learning?

    My passion for experiments and econometrics knowledge are the two things that motivated me towards machine learning. The concept of derivatives, matrix algebra and mathematics attracted me towards deep learning, natural language processing and artificial intelligence

  • What are the key challenges that you encountered in your career?

    I remember taking a continuous backup of R programs in a notepad, as I did not have the luxury of R studio as a platform. Similarly, I learned Python programming without Jupyter Notebook, which is easily available now for anyone. I learned to program a neural network model using Excel in the year 2012, as it was not provided by any statistical software at that time and open source was just taking its shape

  • How has the situation for ML evolved now?

    Now it is no more a challenge, anyone can train the deepest possible learning model on a cloud platform with a few hundred cores. Now going forward the key challenge would be benchmarking existing platforms, libraries and tools and their application in solving various use cases

  • What according to you are the essential tools for programming?

    I would mostly advocate for open source programming languages such as R and Python, though I will not restrict myself to a handful of programming languages. Learning many tools is an added advantage but not necessary, proficiency in at least one programming language is a must

  • What are the books that you have authored?

    As an author, I wrote my first book R – Data Mining Blueprints for beginner level data scientists using R programming language, in the year 2015. In this book I have taken an example-based approach in explaining the data mining algorithms using R. It is currently recommended as a textbook in HSLS centre, University of Pittsburgh, US. It has been translated into Chinese and Spanish as well. Recently my 4th book PyTorch Recipes by Apress (2019) was released. This book covers tensor operations for dynamic graph-based calculations using PyTorch, PyTorch transformations and graph computations for neural networks, supervised and unsupervised machine learning using PyTorch, Work with deep learning algorithms such as CNN and RNN, LSTM models in PyTorch and text processing and NLP.

  • Any words of advice for people who want to make a career in ML.

    Keep Experimenting, till the time you learn the concept.

  • What needs to be done to be a master in this field?

    Tools and programming languages are changing, high-level libraries are replacing base libraries, and advanced computation is eating into the base-level models. Hence there is a constant need to stay updated and upgraded for future challenges. Excelling in AI/ML field requires a lot of patience, constant effort in trying and experimenting with different ways of computation. To be an expert in AI/ML it takes time, a significant amount of time, but a fine start is necessary

  • What will take AI capabilities to the next level?

    More understanding about business and the way business functions, this is necessary to take AI to the next level.

  • What is your plan for the future?

    My future plan would be to do quick experiments on what comes next in AI and ML space and take the early mover advantage in making Leni more intelligent. As an example, capsule networks have emerged as a successor to deep neural networks. There is some traction in the adoption of capsule networks in image-recognition use cases, especially in addressing the drawbacks in CNN (convolutional neural networks) which can further be extended to RNN, LSTM and GRU models.

  • To what extent will AI replace the jobs done by humans?

    No human job can be replaced, AI will replace repetitive tasks, that's why the automation is coming to play

  • What is the driving force behind today's AI progress?

    Algorithms and advanced machines with huge computing powers and advancement in open source software platform are the drivers for AI progress.

  • How do you think AI will impact the future of work?

    low level and repetitive jobs will be replaced by automatic models and algorithms.

  • What skill sets are required to operate AI in an organisation?

    ML, statistics, Linear algebra, Deep learning, mathematics

  • How does the future of AI look 15 years from now?

    more tools, more algorithms, more automatic, and mostly humans will control everything with voice and audio.