The unexpected impact of COVID on a global scale has highlighted the importance of planning to tackle unexpected contingencies. Enterprises that had implemented advanced AI for their business were able to adapt quickly to the changing scenario, as they could use data from various sources to accurately predict the demand-supply patterns.
Having realized the benefits of digitization and AI-based efficacy, traditional industries will also adopt AI and data-based approaches. AI primarily depends on data. With the pandemic, industries are compelled to go digital. This digital transformation foresees a deluge of digital data being available at a relatively low cost, creating new potential.
The pandemic has also forced the education sector to go digital. AI has the potential to augment the learning to a wide spectrum of students with different learning needs. For instance, Edutech platforms are using AI for evaluating the student engagement of their digital content, forming cohorts, providing relevant recommendations to enhance course completion, tracking the retina movements of the students, and so on. Post-COVID, these platforms will increasingly use AI-enabled tools to improve the learning experience.
AI has been a powerful catalyst to expedite the search for candidate molecules for vaccines. Continuing with such similar efforts, AI along with other computational analysis techniques will be crucial for drug discovery and drug remodeling.
During the pandemic, AI is being used to detect facial masks and monitor social distancing protocols. However, after the pandemic recedes, we need to be cautious of smart surveillance being misused for intruding on privacy.
AI tools, developed and enhanced during COVID, helped clinical and biomedical researchers to quickly sort through research literature and tons of experimental/clinical data and find patterns. Going forward, other researchers will also use similar AI tools to speed up their research.
The pandemic also led to a significant increase in the implementation of AI technology in various clinical settings for predicting, diagnosing and managing chronic and mild diseases. Tele-medicine and video consultations enabled medical experts to reach out to rural areas. These tele-medicine platforms, augmented with AI and NLP support, can go a long way in reducing the 'diagnosis to treatment’ time lag.
COVID also exposed the inequalities and biases of our society. Going forward, policy makers, governments, and citizens will use the power of AI to make sure all citizens avail the benefits and facilities with equity and without any discrimination.
PyTorch is an open-source machine learning framework that accelerates the path from research prototyping to production deployment. It supports distributed training, a useful feature that is required for many projects. PyTorch provides two high-level features - Tensor computation (like NumPy) with strong graphics processing units (GPU) acceleration, and deep neural networks built on a tape-based autograd system. It is integrated into Python and designed to be intuitive, linear in thought, easy to use, and debug.
TensorFlow is an end-to-end open-source platform for machine learning (ML). It has a comprehensive, flexible ecosystem of tools, libraries, and community resources that allow researchers to define the state-of-the-art in ML and developers to easily build and deploy ML- powered applications. Developers can also build and train ML models easily using intuitive high-level APIs like Keras, enabling immediate model iteration and easy debugging.
Keras is a high-level deep learning API written in Python, running on top of the machine- learning platform TensorFlow. It was developed to enable fast experimentation, and provides essential abstractions and building blocks for developing and shipping machine learning solutions with high iteration rate. Keras empowers engineers and researchers to take advantage of the scalability and cross-platform capabilities of TensorFlow 2.
Huggingface is a repository and framework for advanced NLP on TensorFlow, PyTorch and Jax. Huggingface provides thousands of pre-trained models to use on texts such as classification, information extraction, question/answers, summarization, translation, text generation and more in over 100 languages. The objective is to make it easier for everyone to use NLP. Transformers provide APIs to quickly download and use the pre-trained models on a given text, modify them on your own datasets and then share them with the community on the model hub.
Ray is a distributed execution framework that enables to scale your applications and leverage machine learning libraries. On top of Ray, there are many libraries for solving problems in ML.
The Ray-Serve library can be used to serve everything from deep learning models built with frameworks like PyTorch, Tensorflow, and Keras, to Scikit-Learn models and arbitrary Python business logic.
MLflow is an open-source platform to manage the ML lifecycle, including experimentation, reproducibility, deployment, and a central model registry. It currently offers four components:
The Kubeflow project aims to enable the deployment of machine learning workflows on Kubernetes that is simple, portable, and scalable. It aims to provide a straightforward way to deploy the best open source systems for ML as well as diverse infrastructures. Kubeflow includes services to create and manage interactive Jupyter notebooks, work locally and deploy them to cloud. Kubeflow's job operator can handle distributed TensorFlow training jobs. It also has support for models serving on various inference optimized servers.
Detectron2, a library from FaceBook AI Research labs (FAIR), provides advanced deep learning-based detection and segmentation algorithms. It supports several computer vision research projects and production applications in Facebook, and includes capabilities such as Mask R-CNN, panoptic segmentation, Dense pose, Cascade R-CNN, rotated bounding boxes, PointRend, among others.
Amazon contact lens is primarily used for sentiment analysis, conversational search, real-time transcript and data redaction. It can be leveraged for service-desk automation and contact-center automation.
This is used for enabling custom voice for service desks and customer service. The communication service is a cloud telephony and omnichannel user experience. It can be leveraged for omnichannel interactions across channels and brand specific engagement with a custom role.
The ES Mega agent helps in building multi-functional conversational bots. CX has a rich interface to configure complex dialogs for multiple conversation topics. It can be used to build complex dialogs often involving multiple conversation topics.
Mindtree has built a comprehensive conversational AI platform called MindFlow for developing conversational applications. It helps organizations design, visualize, build, customize, test, and orchestrate conversational applications at scale. Some of the features in MindFlow include:
RLLib is an open-source library for reinforcement learning that offers both high scalability and a unified API for a variety of applications. RLLib natively supports TensorFlow, TensorFlow Eager, and PyTorch, but most of its internals are framework agnostic.
PyTorch Geometric is a geometric deep-learning extension library for PyTorch. It consists of various methods for deep learning using graphs and other irregular structures, also known as geometric deep learning, from a variety of published papers. In addition, it consists of an easy-to-use mini-batch loader for many small and single giant graphs, multi-GPU support, a large number of common benchmark datasets, and supportive transformations, both for learning on arbitrary graphs as well as on 3D meshes or point clouds.
DGL is an easy-to-use, high performance and scalable Python package for deep learning on graphs. DGL is framework agnostic, that is, if a deep graph model is a component of an end-to-end application, the remaining logic can be implemented in any major framework, such as PyTorch, Apache MXNet or TensorFlow.
PySyft is a Python library for secure and private deep learning. PySyft decouples private data from model training, using federated learning, differential privacy, and encrypted computation [multi-party computation (MPC) and homomorphic encryption (HE)] within the main deep- learning frameworks like PyTorch and TensorFlow.
Opacus is a high-speed library for training PyTorch models with differential privacy (DP) that’s more scalable. Opacus defines a lightweight API by introducing the Privacy Engine abstraction, which takes care of both tracking your privacy budget and working on your model’s gradients.
Spark NLP is a production grade NLP library focusing on healthcare. It enables fast and trainable implementation of advanced biomedical research. It has algorithms for clinical entity recognition and linking, relation extraction, de-identification, and others. It also includes transformer-based medical models and pre-trained models for various medical areas like anatomy, drugs, clinical models, etc.
NeMo is a toolkit for creating conversational AI applications. The toolkit comes with extendable collections of pre-built modules and ready-to-use models for automatic speech recognition (ASR), natural language processing (NLP) and, text-to-speech (TTS). Built for speed, NeMo can utilize NVIDIA's Tensor Cores and scale out training to multiple GPUs and multiple nodes.
Captum is an open source, extensible library for model interpretability built on PyTorch. It supports interpretability of models across modalities including vision, text, and more. Captum provides advanced algorithms, including integrated gradients, to enable researchers and developers to understand the features that contribute to a model’s output. For model developers, Captum can be used to improve and troubleshoot models by identifying different features that contribute to a model’s output.
Fairlearn is an open-source, community-driven project to help data scientists improve the fairness of AI systems. It provides a toolkit of Python-based algorithms for mitigating unfairness in a variety of AI tasks and fairness definitions.
Generative Pre-trained Transformer 3 (GPT3) is large generative neural network model from OpenAI. GPT3 can generate text with a text prompt, and answer questions, write essays, summarize long texts, translate languages, take memos, and even create computer code. It is not openly available and currently, it has limited access to approved customers.
Amazon Connect is an omnichannel cloud contact center that can be used to provide superior customer service. With unified customer data and an ML-based automated agent, and by using NLP capabilities along with voice, IVR, and other capabilities, it enables customization on customer interactions along with cloud-based scaling and self-service through automation, Amazon Connect can be leveraged for service desk and contact center.
Twilio Flex is a programmable cloud contact center platform that provides complete control on deployment, and customized cloud contact centers by overcoming the limitations of SaaS applications. This enables developers to customize and deploy contact centers with Flex tools and use these tools for service desk automation.
Alexa Presentation Language (APL) enables developers to create visual experiences to accompany skills to interact on supported devices such as the Echo Show, Fire TV, some Fire tablets, and others. Animations, graphics, images, slideshows, and video can be included in the visual experience.
Multi-capability skills (MCS) allows developers to extend the built-in support for certain utterances and customize the voice interaction model. For example, support custom utterances, such as "Alexa, ask invocation name what this skill can do?" in the same skill.
InterpretML is an open-source package that incorporates advanced machine learning interpretation techniques in a single package. With this package, you can train interpretable glass-box models and explain black-box systems. InterpretML helps you understand your model's global behavior or understand the reasons behind individual predictions.
JAX is Autograd and XLA, brought together for high-performance machine learning research. JAX can automatically differentiate native Python and NumPy functions. It can differentiate through loops, branches, recursion, and closures, and can take higher order derivatives. It supports reverse-mode differentiation (a.k.a. backpropagation) via grad as well as forward-mode differentiation, and the two can be composed arbitrarily to any order. JAX uses XLA to compile and run your NumPy programs on GPUs and TPUs. Compilation happens under the hood by default, with library calls getting compiled just in time and executed.
JAX also allows you to compile your own Python functions into XLA-optimized kernels using a single-function API. Compilation and automatic differentiation can be composed arbitrarily, so you can process sophisticated algorithms and get maximal performance without leaving Python.
The MONAI framework is an open-source foundation created by Project MONAI. MONAI is a freely available, community-supported, PyTorch-based framework for deep learning in healthcare imaging. It provides domain-optimized foundational capabilities for developing healthcare imaging training workflows in a native PyTorch paradigm. It supports classification and segmentation of 2D and 3D medical imaging data.
EleutherAI-GPT-NEOX is an open-source implementation of model and data parallel auto-regressive GPT3-like generative models. The pre-trained models are available through Huggingface integration.
Pyro is a universal probabilistic programming language (PPL) written in Python and supported by PyTorch at the backend. Pyro enables flexible and expressive deep probabilistic modelling, unifying the best of modern deep learning and Bayesian modelling. It was designed with these key principles: