Machine Learning vs Deep Learning

The superiority of DL approaches compared to other ML algorithms manifests itself when large (huge) quantities of training data are available.

The training of neural networks requires a specialized hardware:

• Before starting a project with DL, you need to ask if the company / lab has the necessary hardware
• Having one or more GPU available is a fundamental factor today (GPU are essential for parallelizing calculations)
• The deeper a network is, the more computational load is introduced

Hardware purchase for DL

With in-house solutions, the company buys the necessary hardware and is the direct owner:

• Pros:
  • Extreme freedom of use of hardware
  • In the long run, it tends to have lower costs
• Cons:
  • Hardware maintenance is required (specialized technicians)
  • Hardware ages quickly
  • For large number of GPUs -> specific server rooms (with high energy consumption)
  • The GPU market is quite expensive and volatile

With external solutions, the hardware is rented through the PaaS paradigm (Cloud).

• Pros:
  • Hardware maintenance is not required
  • No investment over time is required for hardware upgrades
  • Dedicated server rooms are not required, energy consumption is not borne by the company
• Cons:
  • In the long run, it tends to have higher costs
  • Vendor lock-in
  • We do not really know who the owner of the data is
  • Privacy issues

---

Last update: November 30, 2022 15:42:49
Created: November 30, 2022 14:38:50

Authors: Francesca Neri