Neural module networks: A review

Fashandi, Homa

doi:10.1016/j.neucom.2023.126518

Cited by 2 publications

(2 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the field of neural networks some examples are: 1) "Oldschool" attention (Bahdanau et al, 2015;Jain & Wallace, 2019), where attention points to which input tokens are important. 2) Neural Modular Networks (Andreas et al, 2016;Gupta et al, 2020;Fashandi, 2023), which produce a prediction via a sequence of sub-models, each with known behavior. 3) Prototypical Networks (Bien & Tibshirani, 2009;Kim et al, 2014;Chen et al, 2019), which predicts by finding similar training observations.…”

Section: The Intrinsic Paradigmmentioning

confidence: 99%

“…Likewise, Neural Modular Networks produce an executable problem composed of sub-networks, such as find-max-num(filter(find())), which is interpretable (Fashandi, 2023;Andreas et al, 2016;Gupta et al, 2020). However, each sub-networks (find-max-num, filter, find) is itself a black-box model with little guarantee that it operates as intended (Amer & Maul, 2019;Subramanian et al, 2020;Lyu et al, 2022).…”

Section: The Case Against the Intrinsic Paradigmmentioning

confidence: 99%

See 1 more Smart Citation

Post-hoc Interpretability for Neural NLP: A Survey

Madsen¹,

Reddy²,

Chandar

2022

ACM Comput. Surv.

View full text Add to dashboard Cite

Neural networks for NLP are becoming increasingly complex and widespread, and there is a growing concern if these models are responsible to use. Explaining models helps to address the safety and ethical concerns and is essential for accountability. Interpretability serves to provide these explanations in terms that are understandable to humans. Additionally, post-hoc methods provide explanations after a model is learned and are generally model-agnostic. This survey provides a categorization of how recent post-hoc interpretability methods communicate explanations to humans, it discusses each method in-depth, and how they are validated, as the latter is often a common concern.

show abstract

Section: The Intrinsic Paradigmmentioning

confidence: 99%

Section: The Case Against the Intrinsic Paradigmmentioning

confidence: 99%

Post-hoc Interpretability for Neural NLP: A Survey

Madsen¹,

Reddy²,

Chandar

2022

ACM Comput. Surv.

View full text Add to dashboard Cite

show abstract

行動の神経回路制御に関する研究法の技術革新と再解釈

ARAKAWA,

ARAKAWA

2023

Japanese Journal of Animal Psychology

View full text Add to dashboard Cite

Empirical studies on behavior using animal models provide irreplaceable tools for dissecting intricate neural circuits into multiple output modules that concomitantly regulate segmented components of behavior. Recent technical advancement in the manipulation and measurement of neural cell/circuit activities in freely behaving animals allow us to bridge the gap between neural processes and behavior expression. While neural signals are processed on the second timescales, the nature of behavior is time-consuming effort to gain interaction with environmental stimuli for adaptation. Therefore, a bundle of neural activities cannot simply represent entire behavioral processes, but rather segmented behavior components. Hence, we must reconsider how such bundles compose total outputs forming functional set of behavior, and thus, how integrated behavior can be broken into neural modular components, such as sensory detection, element recognition, behavior drive, and action/movement execution, by neural activity timescale units. Our refined protocol on behavioral studies is more effective in terms of determining what we observe in animal behavior and how we interpret neurobehavior relationship for vital contribution to basic and translational studies. Key words：brain and behavior, neural circuits, behavior system, neuropsychology １．問題動物を用いた心理学の研究において，動物が示す行動は主要な測定対象であり，実験事態として用意された環境刺激との相互作用として表出された行動が動物の心的状態を反映していると考える (Kantor, 1970；Levitis et al., 2009 。つまり行動は特定の動きや姿勢として観察され，それを制御しているものを心的状態と仮定している。この心的状態というのは説明変数であり，外的な刺激(Ｓ)に対する行動反応(Ｒ)との関係性を規定する概念的なシステムを想定している。このシステムは，中枢神経系を中心とする分子神経活動の出力を担い，環境刺激への積極的働きかけを行う適応機能を発揮すると考えられる(Mayr, 1997；van der Staay, 2006； Cauchoix & Chaine, 2016) 。この文脈で言えば，行動とは中枢神経系の総体的活動の結果として出力される環境適応作用の表れであり，中枢神経系という機構(メカニズム)に基づいた適応機能を表している(Krakauer et al., 2017) 。メカニズムと機能というのは説明概念のレベルの１)琉球大学大学院医学研究科

show abstract

Neural module networks: A review

Cited by 2 publications

References 28 publications

Post-hoc Interpretability for Neural NLP: A Survey

Post-hoc Interpretability for Neural NLP: A Survey

行動の神経回路制御に関する研究法の技術革新と再解釈

Contact Info

Product

Resources

About