General Synaptics

Safety_Critical_Machine_Learning

What is Third-Wave AI?

The Defense Advanced Research Projects Agency (DARPA) has defined three waves of artificial intelligence:

First Wave AI focused on handcrafted knowledge, logic, and rules.
Second Wave AI brought probabilistic methods, statistical learning and big data.
Third-Wave AI aims to produce contextual adaptation and common sense capabilities.

DARPA proposes that Third-Wave AI systems can understand context, leverage that contextual understanding for common sense reasoning, and adapt to changing contexts. This will enable more natural and intuitive interactions between AI systems and human users.

Key factors are:

More natural language interactions
Improved reasoning skills
Better ability to operate in the open world
Contextual understanding - ability to incorporate and reason about real-world context
Common sense reasoning - making sensible inferences about unfamiliar situations
Adaptability - adjusting to new tasks, users, situations and environments
Intuitive interfaces - more natural language and multimodal interactions

We do research on the following key factors:

Better ability to operate in the open world
Contextual understanding - ability to incorporate and reason about real-world context
Common sense reasoning - making sensible inferences about unfamiliar situations
Adaptability - adjusting to new tasks, users and environments

We design basic Machine Learning Agents which can be interconnected in a complex cognitive system. These algorithms can cooperate in a swarm intelligence context through communication channels that correspond to different information flows relating to different sensory and temporal dimensions. These agents are based on Machine Learning models that have the following characteristics:

Continuous Learning Capability (Plasticity + Stability)
Statistical Consistency Selective Availability
Supervised Learning via Inter-Stream Reinforcement
Adaptability through Unsupervised Learning
Multi Stream Long Term Memory (MS-LTM)
Multi Stream Short Term Memory (MS-STM)
Concept construction in multi-sensory and space-time context

NOT SIMPLE STATISTICAL MULTIMODAL MODELS BUT

COGNITIVE MULTIMODAL MODELS

(TIME AWARE STM/LTM ENTANGLED MODALITIES)

Learning Adaptive Evolving Fuzzy Cognitive Map (LAE-FCM™)

The COGNITIVE INTELLIGENCE is organized as a Learning Adaptive Evolving Fuzzy Cognitive Map (LAE-FCM™)

□ “SENSE” CAN BE A PRIMARY SENSE: SIGHT, HEARING, TOUCH

□ “SENSE” CAN BE ANY STRUCTURED INFORMATION DERIVED FROM THE PRIMARY SENSES: LANGUAGE, TEXT, SHAPES

□ “SENSE” CAN BE ANY POSITIVE OR NEGATIVE EMOTION

□ “SENSE” CAN BE A TARGET OR A STRATEGY TO GET IT AND BOTH ARE A CONCEPT

□ …EVERYTHING IS ENTANGLED IN A SPACE AND TIME DOMAIN…

Emulating intelligence means building a model that replicates stimulus-response associations: this is what current LLM and multimodal AI models do.

Simulating intelligence means trying to reproduce the internal processes that build it: this is what we try to do with our models.

WhoWeAre WhatWeDo Technology

Customers Partners News

General Synaptics

Aerospace_and_Defence_Machine_Learning_Company

VAT NUMBER:_IT02670700992

REA NUMBER: GE-503104

NATO_CAGE_CODE: AK845

Email:_luca.marchese@synaptics.org