I am a postdoctoral researcher at the intersection of machine learning and physics. I have been working at Mila, using deep learning techinques to help find new physics signals at the Large Hadron Collider (LHC) at CERN. I design new algorithms and develop machine learning strategies for LHC physics.

My research spans a wide range of cutting-edge topics, including physics-inspired neural network architectures, anomaly detection for scientific discovery, generative modelling for physics event simulation, and model interpretability and explainability, ensuring a comprehensive understanding of advanced computational techniques and their practical applications.

Prior to my machine learning research career, I was a theoretical physicist working on Supersymmetry (SUSY) phenomenology, precision calculation, and jet physics.

More information can be found in my CV.

  • Check out my GitHub for project-related information: GitHub

Check out my ceramic arts at: TaoCeramics



Energy-based Models for High Energy Physics
Advanced generative modeling for fast event simulation at the Large Hadron Collider
Energy-based models have the natural advantage of flexibility in the form of the energy function. Recently, energy-based models have achieved great success in modeling high-dimensional data in computer vision and natural language processing. In accordance with these signs of progress, we build a versatile energy-based model for High Energy Physics events at the Large Hadron Collider. This framework builds on a powerful generative model and describes higher-order inter-particle interactions. It suits different encoding architectures and builds on implicit generation. As for applicational aspects, it can serve as a powerful parameterized event generator, a generic anomalous signal detector, and an augmented event classifier.

Invariant Representation Driven Classifiers for Anomaly Detection
A method to unify Standard Model jet classifiers and model-independent new physics searches
We leverage representation learning and the inductive bias in neural-net-based Standard Model jet classification tasks, to detect non-QCD signal jets. In establishing the framework for classification-based anomaly detection in jet physics, we demonstrate that, with a well-calibrated and powerful enough feature extractor, a well-trained mass-decorrelated supervised Standard Model neural jet classifier can serve as a strong generic anti-QCD jet tagger for effectively reducing the QCD background. Imposing data-augmented mass-invariance (and thus decoupling the dominant factor) not only facilitates background estimation, but also induces more substructure-aware representation learning. We are able to reach excellent tagging efficiencies for all the test signals considered. In the best case, we reach a background rejection rate of 51 and a significance improvement factor of 3.6 at 50% signal acceptance, with the jet mass decorrelated. This study indicates that supervised Standard Model jet classifiers have great potential in general new physics searches.

Datasets for Machine Learning in High Energy Physics

* Test sets for jet anomaly detection at the LHC

Deep Generative Models for New Physics Searches

Details see our github repo.

Interpretability of Machine Learning Models in High Energy Physics

Using deep neural networks for identifying physics objects at the Large Hadron Collider (LHC) has become a powerful alternative approach in recent years. After successful training of deep neural networks, examining the trained networks not only helps us understand the behaviour of neural networks, but also helps improve the performance of deep learning models through proper interpretation. We take jet tagging problem at the LHC as an example, using recursive neural networks as a starting point, aim at a thorough understanding of the behaviour of the physics-oriented DNNs and the information encoded in the embedding space. We make a comparative study on a series of different jet tagging tasks dominated by different underlying physics. Interesting observations on the latent space are obtained.

Deep Learning in Jet Physics

Being an experimental science, modern particle physics depends heavily on the colliders we can build, and it becomes more difficult to build colliders with higher energy (which really requires more advanced techniques of accelerating particles). Also building bigger colliders requires huge financial support which sometimes may be difficult for governments. While waiting for the financial support and the breakthroughs of hardware techniques supporting new colliders, we still can push the field forward. And the alternative solution might be combing with new techniques in computer sciences and preparing the framework for data analysis for even bigger colliders (the LHeC and the FCC-eh, etc.) which will bring even bigger datasets.
Thus, for this new research branch, the whole ecosystem of high energy physics research will experience a revolution. It requires intimate collaboration from experimentalists, theorists, phenomenologists and computer scientists. And expertise in data science will become basic skill kit for all the categories. Thus in the next few decades, we look forward to transforming the research regime systematically.
Furthermore, the new architectures or other new techniques or theoretical considerations motivated by our research in jet physics/LHC physics would be well-suited for more general practical problems (or other physical science domains), which will bring the high energy physics again interacting with the real life. It will be a win-win regime, in which theory becomes alive and practice has theoretical support.

Jet Physics at the Large Hadron Collider (LHC)

At the LHC, the very high energy produces boosted jets, which makes jet substructure an important topic for LHC Physics. We can use jet substructure to distinguish new physics signals from background events.

Supersymmetry Phenomenology

Supersymmetry is the most promising candidate to solve the "hierarchy problem". After the observation of 125 GeV Higgs boson at the LHC, supersymmetric models have been confronted with severe constraints since it's a little bit difficult to achieve a Higgs boson mass as high as 125 GeV with sparticles staying light around TeV. And along with null results of direct SUSY search, SUSY is somewhat in trouble. Working with Prof. Tianjun Li, I investigated a specific SUSY scenario in which coloured sparticles are heavy enough (several TeV) to give the correct Higgs mass and escape LHC bounds while the electroweak sparticles remaining lighter than TeV. This electroweak SUSY (EWSUSY) scenario fits well with all the experimental constraints. I did global fit for different dark matter categories and also examined the LHC SUSY search bounds on the model parameter space. Results showed that electroweak sector would be very promising for direct SUSY search.
I also examined other SUSY senarios including general NMSSM, EWSUSY in NMSSM, SUSY with Heavy Lightest Supersymmetric Particle (HLSP), etc.

Higher Order Calculation

Except for direct search of Beyond Standard Model signature, new physics also manifest themselves in loop effects. Working with Prof. Dr. Wolfgang Hollik, I have calculated the full next-to-leading order (NLO) electroweak corrections to Higgs-strahlung processes at the LHC for Two-Higgs-Doublet-Model (THDM). By investigating new contributions which are unique in THDM, we can discriminate models from the LHC measurements. And the results show that the NLO electroweak corrections in THDM can be significant due to the non-decoupling effects.This differs from cases in SM and MSSM, leading to a salient feature of THDM at the LHC.


Versatile Energy-Based Models for High Energy Physics
Taoli Cheng and Aaron Courville

Bridging Machine Learning and Sciences: Opportunities and Challenges
Taoli Cheng

Invariant Representation Driven Neural Classifier for Anti-QCD Jet Tagging
Taoli Cheng, Aaron Courville

Variational Autoencoders for Anomalous Jet Tagging
Taoli Cheng, Jean-Fran├žois Arguin, Julien Leissner-Martin, Jacinthe Pilette, Tobias Golling

Interpretability Study on Deep Learning for Jet Physics at the Large Hadron Collider
Taoli Cheng.
Machine Learning and the Physical Sciences workshop (NeurIPS 2019). [arXiv:1911.01872]

Recursive Neural Networks in quark/gluon Tagging, [inspire]
Taoli Cheng.
Comput Softw Big Sci (2018) 2: 3. [arXiv:1711.02633]

Supersymmetry with a Heavy Lightest Supersymmetric Particle, [inspire]
Taoli Cheng, Jinmian Li, Tianjun Li.
Journal of Physics G42 (2015) no.6, 065004. [arXiv:1407.0888]

Electroweak Supersymmetry (EWSUSY) in the NMSSM, [inspire]
Taoli Cheng, Tianjun Li.
Phys.Rev. D88 (2013) 015031. [arXiv:1305.3214]

Natural NMSSM confronting with the LHC7-8, [inspire]
Taoli Cheng, Jinmian Li, Tianjun Li, Qi-Shu Yan.
Phys.Rev. D89 (2014) no.1, 015015. [arXiv:1304.3182]

Toward the Natural and Realistic NMSSM with and without R-Parity, [inspire]
Taoli Cheng, Jinmian Li, Tianjun Li, Xia Wan, You kai Wang, Shou-hua Zhu.

Electroweak Supersymmetry around the Electroweak Scale, [inspire]
Taoli Cheng, Jinmian Li, Tianjun Li, Dimitri V. Nanopoulos, Chunli Tong.
Eur.Phys.J. C73 (2013) 2322. [arXiv:1202.6088].

@import "jekyll-theme-primer";