Learning with Configurable Operators and RL-Based Heuristics

Martínez-Plumed, Fernando; Ferri, Cèsar; Hernández-Orallo, José; Ramírez-Quintana, Marïa José

doi:10.1007/978-3-642-37382-4_1

Cited by 8 publications

(10 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We plan to apply the setting to some other applications, by using the same or other deductive and inductive engines, including rules dealing with numerical or continuous features, such as time or space, and other notions of non-crisp coverage and example certainty, such as probabilities, degrees of truth, fuzzy rules, and so on. We will also keep on with the integration into our declarative learning system 5 (Martínez-Plumed, Ferri, Hernàndez-Orallo, & Ramírez-Quintana, 2013). Furthermore, other cognitive or AI systems, such as decision support systems, can benefit from the application of the metrics introduced here.…”

Section: Discussionmentioning

confidence: 99%

“…The development of a new learning system for knowledge acquisition that is meant to be cumulative is not an easy task. In fact, this research started while developing our system (Martínez-Plumed, Ferri, Hernández-Orallo, & Ramírez-Quintana, 2013, 2014). We were looking for a proper foundation for detailed knowledge assessment metrics and criteria for forgetting.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Knowledge acquisition with forgetting: an incremental and developmental setting

et al. 2015

Self Cite

View full text Add to dashboard Cite

E-mails: {fmartinez,cferri,jorallo,mramirez}@dsic.upv.es AbstractIdentifying the balance between remembering and forgetting is the key to abstraction in the human brain and, therefore, the creation of memories and knowledge. We present an incremental, lifelong view of knowledge acquisition which tries to improve task after task by determining what to keep, consolidate and forget, overcoming the stability-plasticity dilemma. Our framework can combine any rule-based inductive engine (which learns new rules) with a deductive engine (which derives a coverage graph for all rules) and integrates them into a lifelong learner. We rate rules by introducing several metrics through the first adaptation, to our knowledge, of the Minimum Message Length (MML) principle to a coverage graph, a hierarchical assessment structure which handles evidence and rules in a unified way. The metrics are used to forget some of the worst rules and also to consolidate those selected rules that are promoted to the knowledge base. This mechanism is also mirrored by a demotion system. We evaluate the framework with a series of tasks in a chess rule learning domain.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Knowledge acquisition with forgetting: an incremental and developmental setting

et al. 2015

Self Cite

View full text Add to dashboard Cite

show abstract

“…declarative learning system (gErl) [1,2,3,4] that meets several desirable characteristics in terms of expressiveness, comprehensibility and versatility. We have shown that more general systems can be constructed by not only giving power to data and background knowledge representation but also to a flexible operator redefinition and the reuse of heuristics across problems and systems.…”

Section: Extended Abstractmentioning

confidence: 99%

“…Firstly we present a general-purpose There are two internal repositories containing rules and programs which are updated at each learning step by the Rule Generator and the Program Generator processes. The Reinforcement Learning Module is in charge of defining combinations of states and actions (operator applied over a rule) which will be used by the Rule Generator process to select future actions to perform.declarative learning system (gErl) [1,2,3,4] that meets several desirable characteristics in terms of expressiveness, comprehensibility and versatility. We have shown that more general systems can be constructed by not only giving power to data and background knowledge representation but also to a flexible operator redefinition and the reuse of heuristics across problems and systems.…”

mentioning

confidence: 99%

Incremental and developmental perspectives for general-purpose learning systems

Martínez-Plumed

2017

Self Cite

View full text Add to dashboard Cite

The stupefying success of Artificial Intelligence (AI) for specific problems, from recommender systems to self-driving cars, has not yet been matched with a similar progress in general AI systems, coping with a variety of (different) problems. This dissertation deals with the long-standing problem of creating more general AI systems, through the analysis of their development and the evaluation of their cognitive abilities. It presents a declarative general-purpose learning system and a developmental and lifelong approach for knowledge acquisition, consolidation and forgetting. It also analyses the use of the use of more ability-oriented evaluation techniques for AI evaluation and provides further insight for the understanding of the concepts of development and incremental learning in AI systems.Keywords: artificial intelligence, general-purpose learning systems, inductive programming, reinforcement learning, forgetting, task difficulty, cognitive development, evaluation of artificial systems, intelligence tests Extended AbstractIn the light of all the astonishing achievements in recently AI research, it is becoming increasingly clear that creating artificial intelligence is much more than "pattern matching". However, although it would be unfair to deny that some current AI systems exhibit some intelligent behaviour (especially those that incorporate some learning potential), in general terms, most AI research is focused on designing AI systems for a particular functionality or adapted for a specific problem with no intention whatsoever of featuring intelligence. Up to date, the vast majority of the computer models are mindless rule-followers or cleverly written computer program doing statistical calculations and making predictions based on them. However, what it would mean for a computer to behave in an intelligent way? This thesis [9] states that the answer lies in the construction of systems that go beyond task specific scenarios into more general-purpose ones thus able to learn automatically, not pre-programmed or without fixed handcrafted features.Given the above challenge, in the presented dissertation we characterise a series of human intelligence attributes (incremental, developmental and lifelong learning) and cognitive-oriented procedures (memory and forgetting) that, combined with the use of symbolic AI and symbolic learning, have helped us to develop both a general-purpose learning approach as well as a knowledge handling tool. This ambitious issue should, furthermore, pervade the evaluation procedures in AI where systems are usually evaluated in terms of task performance, not really in terms of intelligence. Hence, AI evaluation must necessarily be linked to the purpose of the discipline: general AI systems should require an ability-oriented evaluation in the same way that specialised AI systems should require a task-oriented evaluation.Particularly, and regarding the construction of more general AI approaches, this thesis contributes with a pair of settings for learning and knowledge acquis...

show abstract

“…One of our design considerations is that the approach should work with any inductive engine that is able to generate rules, such as association rule algorithms, IP systems such as gErl [5], or ILP systems such as Progol [6]. Actually, we must also provide a deductive engine for the representation language.…”

Section: Approachmentioning

confidence: 99%