Causality for Machine Learning

Schölkopf, Bernhard

doi:10.48550/arxiv.1911.10500

Cited by 87 publications

(132 citation statements)

References 54 publications

Supporting

Mentioning

130

Contrasting

Order By: Relevance

“…We show that ML classifiers (Logistic regression and LSTM), when used by themselves directly on time-series measurements are dumb to the temporal/ causal-structure in the data. This fact has also been discussed in existing literature [1,2]. When time-series values were directly passed to the classifiers, LR and LSTM, they failed to learn any causal-structure characteristics from the data.…”

Section: Discussion Concluding Remarks and Future Research Directionsmentioning

confidence: 54%

“…What if I had acted in a different way? Machine intelligence is still far away from answering these kind of questions [1,2].…”

Section: Introductionmentioning

confidence: 99%

“…Further, they lack the ability of generalization which involves transfer of learning from one problem to another and learning features that categorize together datasets that are more alike than different. This lack of generalization stems mainly from an inability of these algorithms to learn causal structures [2].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Learning Generalized Causal Structure in Time-series

Kathpalia¹,

Charantimath²,

Nagaraj³

2021

Preprint

View full text Add to dashboard Cite

The science of causality explains/determines 'cause-effect' relationship between the entities of a system by providing mathematical tools for the purpose. In spite of all the success and widespread applications of machine-learning (ML) algorithms, these algorithms are based on statistical learning alone. Currently, they are nowhere close to 'human-like' intelligence as they fail to answer and learn based on the important "Why?" questions. Hence, researchers are attempting to integrate ML with the science of causality. Among the many causal learning issues encountered by ML, one is that these algorithms are dumb to the temporal order or structure in data. In this work we develop a machine learning pipeline based on a recently proposed 'neurochaos' feature learning technique (ChaosFEX feature extractor), that helps us to learn generalized causal-structure in given time-series data.

show abstract

Section: Discussion Concluding Remarks and Future Research Directionsmentioning

confidence: 54%

“…What if I had acted in a different way? Machine intelligence is still far away from answering these kind of questions [1,2].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Generalized Causal Structure in Time-series

Kathpalia¹,

Charantimath²,

Nagaraj³

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Most prior approaches assume that inputs are structured as disentangled variables [6,16,26,36,48,47], which often does not hold in domains with high-dimensional inputs, i.e., images. While Lopez-Paz et al [29] demonstrated the possibility of observational causal discovery from high-dimensional images, combining causal models and representation learning in such domains still remains an open problem [46]. Hence, we instead explore the approach of regularizing a policy that operates on high-dimensional states.…”

Section: Vq-vae Encodermentioning

confidence: 99%

“…In order to address this causal confusion problem, one can consider causal discovery approaches to deduce the cause-effect relationships from observational data [26,48]. However, it is difficult to apply these approaches to domains with high-dimensional inputs, as (i) causal discovery from observational data is impossible in general without certain conditions 3 [38], and (ii) these domains usually do not satisfy the assumption that inputs are structured into random variables connected by a causal graph, e.g., objects in images [29,46]. To address these limitations, de Haan et al [12] recently proposed a method that learns a policy on top of disentangled representations from a β-VAE encoder [19] with random masking, and infers an optimal causal mask during the environment interaction by querying interactive experts [43] or environment returns.…”

Section: Introductionmentioning

confidence: 99%

Object-Aware Regularization for Addressing Causal Confusion in Imitation Learning

Seo¹,

Liu²,

Li³

et al. 2021

Preprint

View full text Add to dashboard Cite

Behavioral cloning has proven to be effective for learning sequential decisionmaking policies from expert demonstrations. However, behavioral cloning often suffers from the causal confusion problem where a policy relies on the noticeable effect of expert actions due to the strong correlation but not the cause we desire. This paper presents Object-aware REgularizatiOn (OREO), a simple technique that regularizes an imitation policy in an object-aware manner. Our main idea is to encourage a policy to uniformly attend to all semantic objects, in order to prevent the policy from exploiting nuisance variables strongly correlated with expert actions. To this end, we introduce a two-stage approach: (a) we extract semantic objects from images by utilizing discrete codes from a vector-quantized variational autoencoder, and (b) we randomly drop the units that share the same discrete code together, i.e., masking out semantic objects. Our experiments demonstrate that OREO significantly improves the performance of behavioral cloning, outperforming various other regularization and causality-based methods on a variety of Atari environments and a self-driving CARLA environment. We also show that our method even outperforms inverse reinforcement learning methods trained with a considerable amount of environment interaction.

show abstract

AI models and the future of genomic research and medicine: True sons of knowledge?

et al. 2021

View full text Add to dashboard Cite

The increasing availability of large‐scale, complex data has made research into how human genomes determine physiology in health and disease, as well as its application to drug development and medicine, an attractive field for artificial intelligence (AI) approaches. Looking at recent developments, we explore how such approaches interconnect and may conflict with needs for and notions of causal knowledge in molecular genetics and genomic medicine. We provide reasons to suggest that—while capable of generating predictive knowledge at unprecedented pace and scale—if and how these approaches will be integrated with prevailing causal concepts will not only determine the future of scientific understanding and self‐conceptions in these fields. But these questions will also be key to develop differentiated policies, such as for education and regulation, in order to harness societal benefits of AI for genomic research and medicine.

show abstract

Causality for Machine Learning

Cited by 87 publications

References 54 publications

Learning Generalized Causal Structure in Time-series

Learning Generalized Causal Structure in Time-series

Object-Aware Regularization for Addressing Causal Confusion in Imitation Learning

AI models and the future of genomic research and medicine: True sons of knowledge?

Contact Info

Product

Resources

About