• Pages
01 Home
02 Introduction
03 Executive Summary
04 Overview of TechnoVision
05 You Experience
06 Experience²
07 Me Myself and My Metaverse
08 No Friction
09 I Feel for You
10 My Own Private Avatar
11 We Collaborate
12 Fluid Workforce
13 The Team is the Canvas
14 Taken by Tokens
15 Your Business is a Mesh
16 It’s All Connected
17 Thriving on Data
18 Data Sharing is Caring
19 Power to the People
20 Data Apart Together
21 Era of Algorithms
22 Creative Machine
23 Process on the Fly
24 Process is Mine Mine Mine
25 Rock, Robot Rock
26 Silo Busters
27 Can’t Touch This
28 Augmented Me
29 Applications Unleashed
30 Kondo My Portfolio
31 Honey, I Shrunk the Applications
32 When Code goes Low
33 Mesh Up Your Apps
34 Apps ❤️AI
35 Invisible Infostructure
36 Lord of the clouds
37 Crouching Tiger, Hidden Container
38 Simply the Edge
39 Ops, AI did it Again
40 Silence of The Servers
41 Balance by Design
42 Technologyϵ϶Business
43 Adapt First
44 With Open Arms
45 Do Well, Do Good
46 Trust Thrust
47 IQ CQ EQ Up
48 No Hands On Deck
49 A Few More Things
50 Further Research and the Team

ERA OF ALGORITHMS

Challenge everything you’ve tried so far: the next-generation AI algorithms bring brand-new, awesome ways to solve problems, innovate, and bring out the very best in humans

The age of guessing is over. The era of algorithms is here. Much of the current love for AI comes from deep learning on neural networks. These are brute force, pattern recognition machines that – if provided with plenty of training data – can go where the more traditional data science cannot. Deep learning can be combined with other technology-enabled tactics, such as reinforcement learning, to provide even more unparalleled problem-solving power. Building these algorithms is a specialized, energy-consuming task, but off-the-shelf algorithms and models provide sensible, sustainable alternatives.

Padmashree Shagrithaya Expert in Residence

WHAT

  • Many current advances in AI are thanks to machine learning models on neural networks, detecting and classifying features through multiple layers of raw input.
  • With abundant training data as an input, neural networks may recognize patterns much more effectively than traditional (statistical) data science approaches.
  • Advances in the ability to collect, store, and access training data, plus powerful graphical processing units (such as GPUs), have been instrumental to its success.
  • Reinforcement learning uses an action/reward approach to learn from interactions. This creates additional AI power in areas such as robotics, scheduling, and gaming.
  • Training AI models consumes a lot of energy, but the resulting models can optimize energy consumption of a variety of business activities, providing a net gain scenario.
  • Training AI models also requires highly specialized skills, but low-code AutoML (Automated Machine Learning) tools bring AI algorithms to a much wider audience.
  • Approaches such as TinyML offload trained AI models to even the smallest of edge devices, relieving energy-consuming central facilities.

USE

  • Microsoft developed an image-captioning algorithm that exceeds human accuracy in identifying objects, but also more precisely describes the relationship between them. It is incorporated within its assistant app for the visually impaired, “Seeing AI.”
  • A life science research team developed a reinforcement learning-based AI pancreas that calculates the amount of insulin needed for a diabetic patient and injects it automatically. It is hailed as “autonomous driving for the medical industry.”
  • Tokio Marine uses deep learning computer vision to auto process damage insurance, analyzing photos of damaged vehicles and providing recommendations on repair options, paint and blend operations, and the expected number of labor hours.

IMPACT

  • Solving problems that were deemed impossible to solve – or insufficiently successful – with more classic data science approaches.
  • Creating powerful, self-learning, and self-optimizing autonomous systems decreases the need for scarce onsite human resources.
  • Extracting more value out of (historical) data by turning them into powerful AI algorithms that can be monetized externally.
  • Acquiring access to next-generation AI algorithms without the need for scarce, specialized resources – through AutoML, low-code AI, and off-the-shelf models.
  • Lowering the energy consumption of large, central AI computing environments through downloading trained “inference” models to edge devices.
  • Using superior AI algorithms for optimizing scare natural and human resources, to battle climate change, and in general contribute to better societal futures.

TECH

  • Deep learning/neural networks: TensorFlow, Microsoft Cognitive Toolkit, Theano, MXNet, Keras, Chainer, PyTorch, Gluon, Horovod, AWS Deep Learning, Caffe, Deeplearning4j, PlaidML, OpenAI GPT-3
  • Reinforcement learning: AWS DeepRacer, Facebook Horizon, Gym on OpenAI, Microsoft Project Malmo, Google Dopamine, RLLib via Ray Project, Tensorforce, Reinforcement Learning Coach by Intel, MAgent, Tensorflow Agents, SLM Lab, DeeR
  • AI infrastructure accelerators: NVIDIA Deep Learning, AWS Deep Learning AMIs, Google Cloud TPU, Intel AI and Neural Compute Stick, Apple Neural Engine, Qualcomm Cloud AI100, IBM Watson Machine Learning Accelerator, Inference Engine by FWDNXT, ALVEO, tinyML
  • AutoML and low-code AI: Microsoft Project Bonsai, Google Vertex AI, DataRobot, Microsoft Azure AutoML

Find Out More

Visit our website to delve into Technovision trends and resources:

Find out more

TechnoVision Report

To access and download the TechnoVision 2022 report as a PDF, click here:

Download Report

Contact Us

If you would like to know more about TechnoVision from Capgemini, or to speak to us about any aspects of this report, click here:

Contact Us