Knowledge is constantly revised (evolves) as new pieces of information is made available over time. We term it “knowledge augmentation”. Hence it is feasible to achieve knowledge augmentation via incremental learning.
2. ● Background
● Definition
● Approach to ML
● Challenges
● Algorithm
● K nearest neighbor
● SVM
● Neural network
● Application
● Code &Result
● Conclusion
Context
3. Learn, Retain , use knowledge over an extended period of time
New piece of
information
MODEL
● Knowledge constantly revised
● knowledge augmentation via
incremental learning
4. ● Incremental learning refers to the situation of
continuous model adaptation based on a constantly
arriving data stream
● Incremental learning algorithms can constantly learn
new information from new samples when compared with
conventional machine learning. It also preserves most of
the knowledge. It processes big data by sequential
treatment.
Definition
5. Classical approach of ML
● Knowledge is not retained and accumulated.
● Learning performed without considering the knowledge
from the previous learnt task
● Data is given priority
● Static data
DATASET -
1 TASK-1
DATASET -
2
DATASET -
n
TASK-2
TASK-n
6. Incremental learning algorithm
● store new information when available
● handle unlabelled data
● learn incrementally
● update changes in the concept etc.
7. K-NEAREST NEIGHBOUR
● classifying cases based on their closeness in problem
space
● Assign a class based on how close its values of certain
attributes are to the values of attributes from previously
seen cases.
● Learns incrementally
8. Incremental SVM
● fast, numerically stable and robust algorithm
● work effectively with limited resources
● perform active learning in dynamic environment
● Drawback: huge memory requirement
9. Incremental neural network
● iCaRL : add new classes based on learning from
previous classes.
● A subset of training samples from previous classes is
stored.
● The size of the subset is kept constant. As new classes
arrive, some examples from old classes are removed.
10. challenges
● Concept drift
● Data not available, instances are available
● Limited memory resources
● Stability - plasticity dilemma
17. Result
40000000 41000000
Conversion took 55.11772918701172 s
Mean Absolute Error: 1.9487526976244665
Operation 1 took 549.3371121883392
41000000 42000000
Conversion took 52.50961995124817 s
Mean Absolute Error: 1.8803513388788844
Operation 2 took 526.2503442764282
18. Conclusion
● Incremental learning, as an exploration hotspot in the field of
AI, has drawn an ever increasing number of scientists to
investigate and create it.
● The way of extension of gradual learning can be considered
as per the development pattern of profound learning.
● Gradual learning is a system of AI where the learning cycle
happens as new models are produced and what has been
discovered is altered by the new example(s).
Knowledge is constantly revised (evolves) as new pieces of information is made available over time. We term it “knowledge augmentation”. Hence it is feasible to achieve knowledge augmentation via incremental learning.