Offline ai in mit app (using Artifical Intelligence)

MIT App Inventor Help
1

How to make an offline ai . Please give suggestions

2
3

No i want to make an offline ai with good amount paramaters

4

Means?

5

What is AI in this context?
App Inventor?
Artificial Intelligence?
All inclusive?
Aviation Industry?
Assistant Instructor?

Taifun

6

Artificial intelligence

7

To make an app using App Inventor using Artificial Intelligence the app MUST connect to the Internet to make use of the Artificial Intelligence model.

What you want to do offline is not possible. :frowning: for any significant usage.

What you can do is mimic artificial intelligence with something like this

8

Here is what Gemini is suggesting...
Search for Tensorflow, PyTorch and ML Kit here in the community

Taifun

Yes, it is absolutely possible to create an AI app for Android that runs offline! This is a growing area known as Edge AI or On-device AI.

Here's why and how:
Why Offline AI on Android?

  • Privacy: User data stays on the device, never sent to external servers. This is crucial for sensitive information.
  • Low Latency: No internet round-trip means faster responses, essential for real-time applications (e.g., live image processing).
  • No Internet Required: Works anywhere, anytime, regardless of network connectivity (e.g., in remote areas, on planes, in subways).
  • Reduced Costs: No server costs for inference.
  • Power Efficiency: Optimized models can run efficiently on mobile hardware, consuming less power than continuous network usage.

How is it done?
The key is to optimize and deploy AI models that are small enough and efficient enough to run directly on the mobile device's hardware. Here are the primary frameworks and tools used:

  • TensorFlow Lite (now part of LiteRT / Google AI Edge):
    • This is Google's official framework for deploying machine learning models on mobile and edge devices.
    • You train your model (e.g., in TensorFlow or PyTorch) on a powerful machine, and then convert it to the highly optimized .tflite format.
    • TensorFlow Lite provides an interpreter that runs these models efficiently on Android, leveraging the device's CPU, GPU, or dedicated AI accelerators (NPUs).
    • Google is heavily investing in Google AI Edge and LiteRT, which are the evolution of TensorFlow Lite, offering more advanced capabilities for on-device AI, including generative AI models like Gemini Nano.
  • PyTorch Mobile (now part of ExecuTorch):
    • PyTorch also offers solutions for mobile deployment. You can convert PyTorch models to the TorchScript format, which can then be run on Android devices.
    • Similar to TensorFlow Lite, it provides a runtime environment optimized for mobile.
    • ExecuTorch is the new runtime for PyTorch on edge devices, aiming for even greater efficiency and broader platform support.
  • ML Kit (Google):
    • ML Kit offers ready-to-use APIs for common machine learning tasks (like text recognition, face detection, image labeling, smart reply, object detection). Many of these models can run completely on-device.
    • It's a great option if your AI task falls within the scope of its pre-built functionalities, as it simplifies development significantly.
  • Custom Solutions / Other Frameworks:
    • For highly specialized or experimental use cases, developers might use other libraries or even implement custom inference engines, though this is much more complex.
    • Qualcomm AI Hub is another platform that provides tools for optimizing and deploying AI models on Qualcomm-powered Android devices.
      Examples of Offline AI Apps on Android:
  • Image Recognition: Identifying objects, animals, or plants in photos without an internet connection.
  • Natural Language Processing (NLP):
    • Offline chatbots or language models (like the new Google AI Edge Gallery app or "Jan" that run large language models (LLMs) locally).
    • Text summarization, translation, or sentiment analysis directly on the device.
    • Smart replies in messaging apps.
  • Audio Processing: Voice commands, speech-to-text, or sound classification.
  • Object Detection/Tracking: Used in augmented reality (AR) or accessibility apps.
  • Recommendation Systems: Providing personalized recommendations based on local user data.

Key Considerations for Developing Offline AI Apps:

  • Model Size: Mobile devices have limited storage. Models need to be highly compressed (e.g., using quantization or pruning).
  • Computational Resources: Mobile CPUs, GPUs, and NPUs have less power than cloud servers. Models must be optimized for efficient inference.
  • Battery Consumption: Running complex AI models can drain battery life, so efficiency is paramount.
  • Device Compatibility: Performance can vary significantly between different Android devices with varying hardware.

In conclusion, not only is it possible to create offline AI apps on Android, but it's also a rapidly evolving field with significant advancements driven by frameworks like TensorFlow Lite/LiteRT and PyTorch Mobile/ExecuTorch, enabling a new generation of privacy-preserving and highly responsive AI experiences.

9

are you familiar with these options described here? (this is grok's response to my prompt "what's the product from gemini/google that is like an offline AI in your own android?"

https://x.com/i/grok/share/vwvM3TjO5oEiGoqtTKIMuePl3

i've actually installed AI edge gallery (Taifun mentions it above) to my galaxy s20 fe 5g and it works! i installed it as a standby but i hardly use it because i'm almost always near a wi-fi connection. i've tried it to test simple stuff like - write a poem, what's the weather, solve this , chat with me,etc. and they all work as expected.

my only concern is that nce it's not online i suppose i have to upgrade it occassonally.

1 Like