Balancing Familiarity and Curiosity in Data Exploration with Deep Reinforcement Learning

Personnaz, Aurélien; Amer-Yahia, Sihem; Berti‐Équille, Laure; Fabricius, Maximilian; Subramanian, Srividya

doi:10.1145/3464509.3464884

Cited by 6 publications

(15 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We find that the traditional drilldown and roll-up operators are not suffice for finding useful summaries, especially in SDSS that requires more expressive operators to cover the variety of galaxy types. We also ran an experiment to validate our reward function and found that it outperforms baseline DRL with familiarity and curiosity [37]. Finally, an investigation with two domain expert astronomers who are familiar with SDSS revealed the benefit of using partial guidance for summarization.…”

Section: Introductionmentioning

confidence: 93%

“…ML for data exploration: Recent work suggested to automate data exploration using Reinforcement Learning [9,37,38,44]. EDA4Sum adopts a similar approach to provide guidance to users with no need for training data.…”

Section: Related Workmentioning

confidence: 99%

“…EDA4Sum adopts a similar approach to provide guidance to users with no need for training data. The logic of our RLSum algorithm is based on the system presented in [37,38] that guides users in finding items of interest in large datasets. In this system, the process is driven by data familiarity and curiosity.…”

Section: Related Workmentioning

confidence: 99%

“…Following recent approaches that were proven to be useful for EDA [10,37,38,44], we present a Deep Reinforcement Learning solution to find a high-utility summarization pipeline.…”

Section: Rlsummentioning

confidence: 99%

“…Top1Sum is a greedy algorithm that chooses to return the highest utility summary at each subsequent step. Following existing work in EDA [19,37,44] we investigate the applicability of Deep Reinforcement Learning to summarization. We design RLSum, an adaptation of Deep Reinforcement Learning to simulate an agent that learns an end-to-end summarization policy as a sequence of EDA operators that yield the highest reward.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Guided Exploration of Data Summaries

Youngmann¹,

Amer-Yahia²,

Personnaz³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Data summarization is the process of producing interpretable and representative subsets of an input dataset. It is usually performed following a one-shot process with the purpose of finding the best summary. A useful summary contains 𝑘 individually uniform sets that are collectively diverse to be representative. Uniformity addresses interpretability and diversity addresses representativity. Finding such as summary is a difficult task when data is highly diverse and large. We examine the applicability of Exploratory Data Analysis (EDA) to data summarization and formalize Eda4Sum, the problem of guided exploration of data summaries that seeks to sequentially produce connected summaries with the goal of maximizing their cumulative utility. Eda4Sum generalizes one-shot summarization. We propose to solve it with one of two approaches: (i) Top1Sum that chooses the most useful summary at each step; (ii) RLSum that trains a policy with Deep Reinforcement Learning that rewards an agent for finding a diverse and new collection of uniform sets at each step. We compare these approaches with one-shot summarization and top-performing EDA solutions. We run extensive experiments on three large datasets. Our results demonstrate the superiority of our approaches for summarizing very large data, and the need to provide guidance to domain experts.

show abstract

Section: Introductionmentioning

confidence: 93%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

“…Following recent approaches that were proven to be useful for EDA [10,37,38,44], we present a Deep Reinforcement Learning solution to find a high-utility summarization pipeline.…”

Section: Rlsummentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations