Deep learning for the detection of γ-ray sources:
Bridging the gap between training simulations and real telescope observations using unsupervised domain adaptation

GammaLearn update status

Author Under the supervision of
Michaël Dell'aiera (LAPP, LISTIC) Thomas Vuillaume (LAPP)
Alexandre Benoit (LISTIC)
LST meeting (Prague) - 21th May 2024
dellaiera.michael@gmail.com

Introduction

Contextualisation

**[GammaLearn](https://purl.org/gammalearn)** * Fosters innovative methods in AI for CTA * Evaluate the added value of deep learning

Fig. The detection workflow
Fig. γ-PhysNet (Jacquemont et al.)

**[Main results (Published)](https://arxiv.org/abs/2108.04130)** * Outperforms Hillas+RF on MC and on real data in controlled environment * But performances on real data could be improved → Domain adaptation

Sensitivity of γ-PhysNet to NSB

Fig. Impact of NSB. Poisson rate difference δλ=λ_train-λ_test

Simulations and real data discrepencies

**Simulations are approximations of the reality**

Fig. Poisson rate λ(t) as a function of time t. Variation of NSB in Crab runs.

* Poisson approximation * Independent pixels

Fig. Stars in the FoV, dysfunctioning pixels

* Some pixels are correlated

Data adaptation (or data tuning)

**Modify the simulations to fit the acquisitions**

Fig. NSB distributions (Simulations, Crab, Markarian, and data adaptation)

Train

Test
Labelled Labelled
MC+P(λ)

ratio=50%/50% 		MC+P(λ)

ratio=50%/50%

Tab. Dataset composition

Results with data adaptation on Crab (real data)

Multi-modality

γ-PhysNet-CBN (Conditional Batch Norm)

**Modify the model to make it robust to noise**

Fig. γ-PhysNet-CBN architecture
Fig. BN module
Fig. CBN module

Setup

Fig. NSB (Simulations, Crab, Markarian, and data adaptation)

Train

Test
Labelled Labelled
MC+P(λ~Uniform())

ratio=50%/50% 		MC+P(λ)

ratio=50%/50%

Tab. Dataset composition

Results with multi-modality on simulations

Fig. Poisson rate difference δλ=λ_train-λ_test, with λ_train~Uniform(0,1)

Results with multi-modality on Crab (real data)

Multi-modality

Fig. γ-PhysNet-CBN architecture

Domain adaptation

Domain adaptation

**[Domain adaptation](https://arxiv.org/abs/2009.00155): Set of algorithms and techniques to reduce domain discrepancies**

* Take into account unknown differences between the source (labelled, simulations) and target (unlabelled, real data) domains * Somehow include unlabelled real data in the training (→ label shift) * Selection, implementation and validation of [DANN](https://arxiv.org/abs/1505.07818) (focus of this talk), [DeepJDOT](https://arxiv.org/abs/1803.10081), [DeepCORAL](https://arxiv.org/abs/1607.01719)

Fig. Domain confusion in the feature space

Comparative study and validation

* Traditional deep learning dataset (digits) * Most of the difficulty resides in the optimization of the model (conflicting gradients) * Automatic determination of the loss coefficients * Gradient weighting strategy * Best performance using DANN paired with [Uncertainy Weighting](https://arxiv.org/abs/1705.07115) → **Paper #1**: Method-oriented with results on MC (will be send soon for your review)

Fig. Multi-task balancing

γ-PhysNet-DANN (Domain Adversarial Neural Network)

**Modify the model to make it domain agnostic**

Fig. γ-PhysNet-DANN architecture

Application of γ-PhysNet-DANN to simulations

Fig. Results on simulations (NSB, w/o label shift)(Published)

Conditional domain adaptation

Fig. Domain adaptation
Fig. Conditional domain adaptation

Setup

Fig. Conditional domain adaptation

Train

Test
Source
Labelled		 Target
Unlabelled
Unlabelled
MC

ratio=50%/50% 		MC+P(λ)

ratio~0.1%/99.9%
(Label shift) 		MC+P(λ)

ratio=50%/50%

Tab. Dataset composition

Results with domain adaptation on simulations

Fig. γ-PhysNet-DANN vs γ-PhysNet-CDANN

Results with domain adaptation on Crab (real data)

Conclusion and perspectives

Conclusion & Perspectives

  • Comparison of Data adaptation, Multi-modality, Domain adaptation to solve simulations vs real data discreprency
    • Tested on simulations
    • Tested on real data (Crab), both moonlight and no moonlight conditions
  • Standard analysis and γ-PhysNet strongly affected by moonlight
  • Data adaptation and multi-modality increase the performance in degraded conditions
  • The benefits of domain adaptation are not well established yet
    • Tuning accurately the MC to data NSB is still the best performing approach
    • Domain adaptation can match the best performance on tuned simulations (but more variability)
  • γ-PhysNet-CBN with pedestal image conditioning
  • Two articles to be published:
    • Paper #1: Method-oriented with results on MC (will be send soon for your review)
    • Paper #2: Crab analysis
  • My PhD ends September 30th

Acknowledgments

- This project is supported by the facilities offered by the Univ. Savoie Mont Blanc - CNRS/IN2P3 MUST computing center - This project was granted access to the HPC resources of IDRIS under the allocation 2020-AD011011577 made by GENCI - This project is supported by the computing and data processing ressources from the CNRS/IN2P3 Computing Center (Lyon - France) - We gratefully acknowledge the support of the NVIDIA Corporation with the donation of one NVIDIA P6000 GPU for this research. - We gratefully acknowledge financial support from the agencies and organizations listed [here](https://www.cta-observatory.org/consortium\_acknowledgment). - This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 653477 - This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 824064