Cognitively-Inspired Model for Incremental Learning Using a Few Examples

Ayub, Ali; Wagner, Alan R.

doi:10.1109/cvprw50498.2020.00119

Cited by 39 publications

(31 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As the combined datasets were filtered to have an even number of rhyme and non-rhyme examples, the respective models were more focused and able to replicate the targets (the original song lyrics) more accurately. This reflects a general principle of transfer learning: that fine-tuning on too much data can be detrimental to a model's performance [2].…”

Section: Metrics Resultsmentioning

confidence: 97%

Say What? Collaborative Pop Lyric Generation Using Multitask Transfer Learning

Naveen

Gummadi

Bhethanabotla

et al. 2021

Proceedings of the 9th International Conference on Human-Agent Interaction

View full text Add to dashboard Cite

Lyric generation is a popular sub-field of natural language generation that has seen growth in recent years. Pop lyrics are of unique interest due to the genre's unique style and content, in addition to the high level of collaboration that goes on behind the scenes in the professional pop songwriting process. In this paper, we present a collaborative line-level lyric generation system that utilizes transfer-learning via the T5 transformer model, which, till date, has not been used to generate pop lyrics. By working and communicating directly with professional songwriters, we develop a model that is able to learn lyrical and stylistic tasks like rhyming, matching line beat requirements, and ending lines with specific target words. Our approach compares favorably to existing methods for multiple datasets and yields positive results from our online studies and interviews with industry songwriters. CCS CONCEPTS• Human-centered computing → Collaborative and social computing; • Applied computing → Sound and music computing; • Computing methodologies → Natural language processing; Transfer learning.

show abstract

Section: Metrics Resultsmentioning

confidence: 97%

Say What? Collaborative Pop Lyric Generation Using Multitask Transfer Learning

Naveen

Gummadi

Bhethanabotla

et al. 2021

Proceedings of the 9th International Conference on Human-Agent Interaction

View full text Add to dashboard Cite

show abstract

“…Towards the goal of creating continually learning robots, the first project in my PhD was focused on the few-shot incremental learning problem (FSIL), in which the robot learns continually from a small number of object examples provided by a human. I developed a novel approach termed Centroid-Based Concept Learning (CBCL) to tackle this problem [3]. CBCL's classification accuracy was significantly higher than the State-of-the-art (SOTA) incremental learning approaches on benchmark datasets (Table 1).…”

Section: Past Current and Future Workmentioning

confidence: 99%

Continual Learning of Visual Concepts for Robots through Limited Supervision

Ayub

Wagner

2021

Companion of the 2021 ACM/IEEE International Conference on Human-Robot Interaction

Self Cite

View full text Add to dashboard Cite

For many real-world robotics applications, robots need to continually adapt and learn new concepts. Further, robots need to learn through limited data because of scarcity of labeled data in the realworld environments. To this end, my research focuses on developing robots that continually learn in dynamic unseen environments/scenarios, learn from limited human supervision, remember previously learned knowledge and use that knowledge to learn new concepts. I develop machine learning models that not only produce State-of-the-results on benchmark datasets but also allow robots to learn new objects and scenes in unconstrained environments which lead to a variety of novel robotics applications. INTRODUCTIONContinual adaptation and learning through limited data is the hallmark of human intelligence. Humans continue to learn new concepts over their lifetime without the need to relearn most previous concepts. With robots becoming an integral part of our society, they must also continue to learn over their lifetime to adapt to the ever-changing environments. Further, in real-world applications, robots do not have access to a large amount of labeled data since it is impractical for human users to provide hundreds of examples to the robot. Thus, robots must learn using a small amount of data through limited human supervision. The long-term goal of my research is to develop autonomous robots for everyday environments where they can learn over their lifetime and use the learned knowledge to assist humans in their daily lives.Creating robots that continually learn is a challenging problem. Deep learning is widely used to address many robot learning tasks, yet deep learning suffers from a phenomenon called catastrophic forgetting when learning continually. Catastrophic forgetting occurs when continually training a model (neural network) to recognize new classes, the model forgets the previously learned classes and the overall classification accuracy decreases. One way to address this problem is by storing the complete data of the previously learned classes. However, storing data of the previous classes requires a huge memory when learning new classes continually. Robots, on the other hand, have limited on-board memory available, hence they cannot keep storing high-dimensional images of Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s).

show abstract

“…Further, their approach requires the complete training set of each class and cannot be applied for few-shot incremental learning. An extension of [15] is presented in [18] for incremental object learning for a robot using a combination of a pre-trained feature extractor and NCM classifier which has been shown to be significantly inferior to other incremental learning approaches [3], [5].…”

Section: B Incremental Learning Applications In Roboticsmentioning

confidence: 99%

“…This paper builds from our prior research on incremental learning [5] by evaluating a practical application of few-shot incremental learning in which a robot is taught novel object classes incrementally using a small set of visual examples provided by a human. An n-shot incremental learner (where n is usually 1,5, or 10) recognizes objects but is only trained on n examples per class for k classes per increment.…”

Section: Introductionmentioning

confidence: 99%

“…An n-shot incremental learner (where n is usually 1,5, or 10) recognizes objects but is only trained on n examples per class for k classes per increment. We use our centroid-based concept learning (CBCL) approach proposed in [5], [6] which achieves state-of-the-art performance on incremental learning and scene classification benchmarks and can be used for few-shot incremental learning of objects for a table cleaning task. First, the robot is taught classes of objects to clear from a few examples provided by a human incrementally.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Tell me what this is: Few-Shot Incremental Object Learning by a Robot

Ayub

Wagner

2020

2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Self Cite

View full text Add to dashboard Cite

For many applications, robots will need to be incrementally trained to recognize the specific objects needed for an application. This paper presents a practical system for incrementally training a robot to recognize different object categories using only a small set of visual examples provided by a human. The paper uses a recently developed state-of-theart method for few-shot incremental learning of objects. After learning the object classes incrementally, the robot performs a table cleaning task organizing objects into categories specified by the human. We also demonstrate the system's ability to learn arrangements of objects and predict missing or incorrectly placed objects. Experimental evaluations demonstrate that our approach achieves nearly the same performance as a system trained with all examples at one time (batch training), which constitutes a theoretical upper bound.

show abstract

Cognitively-Inspired Model for Incremental Learning Using a Few Examples

Cited by 39 publications

References 36 publications

Say What? Collaborative Pop Lyric Generation Using Multitask Transfer Learning

Say What? Collaborative Pop Lyric Generation Using Multitask Transfer Learning

Continual Learning of Visual Concepts for Robots through Limited Supervision

Tell me what this is: Few-Shot Incremental Object Learning by a Robot

Contact Info

Product

Resources

About