From my work at VMware research, I am interested in both applied and theoretical research in the domains of deep learning and generative AI. Though my work has been mostly of an applied nature since I started my work at VMware, I love finding problems that can benefit from theoretical insights. My interests are diverse.

• LLM performance in multi-agent settings.

  • What helps or hinders the performance of an LLM multi-agents system, according to some specific goal the system is intended to pursue? Do biases play a role? Do some unforeseen emergent behavior occurs? 

  • How to ensure a group of agents behave collectively as expected?

  • How to incentivize collaboration among agents? 

• Enhancing the "reasoning" and "planning" capabilities of LLMs. 

• Trustworthiness and security concerns in LLMs and systems of LLMs.

• Alternatives to transformer-based language models, such as state-space models.

• Diffusion Models: theoretical insights, scalability, application on different domains and systems, etc.

• Alternative Gen AI: state-space models, flow matching, etc.

From my postdoc work, I am interest in theoretical research that: 1) helps us better understand the empirical success of machine learning, and 2) helps us design better machine learning algorithms with theoretical guarantees.

• How can multiple reinforcement learning agents be used to benefit the accuracy or efficiency of RL algorithms that otherwise would have been run using a single agent? 

• How can we formally justify the use of heuristics that perform well in multi-agent RL solutions? 

• What theoretical insights and guarantees can we provide on the training and generalization of different deep learning models (neural networks, LLMs, neural operators, etc.)? How does this change depending on the optimizer being used, as well as any normalization procedure? 

• How do different hyperparameters and initializations schemes affect the interplay among training, generalization and robustness?

From my PhD studies, I am generally interested in the areas of network systems and/or multi-agent systems. Throughout my past work, I have used concepts from game theory, optimization theory and mathematical sociology. Some of my works are summarized in the abstract of a talk (link) I gave at UIUC.

• Modeling of social processes: Evolution of positive and negative interpersonal relationships over a social network (I have worked on social balance theory). Diffusion of opinion dynamics over graphs and hypergraphs. I have used tools from dynamical systems and game theory.

• Stability tools with applications to multi-agent or network systems. My past work has used a strong stability concept for dynamical systems called contraction theory for such purpose. 

• Distributed optimization.

• Epidemic models.

• Distributionally robust optimization.

• Computation and application of optimal transport and Wasserstein barycenters.

Throughout my research career, I have collaborated with faculty and/or post-docs from Sociology, Mechanical Engineering, Statistics, and Computer Science. 

Other Interests:

I am happy to continue working on different problems related to all the aforementioned research areas -- including the ones I have worked in the past. Other areas I would love to explore are multi-modal ML systems, computational economics (e.g., mechanism design, auctions, etc.), and robotics (e.g., learning in robotics, embodied intelligence, smart motion planning, multi-robot systems, etc.). 

Preprints:

• P. Cisneros-Velarde, B. Shrimali, A. Banerjee, "Optimization for Neural Operators can Benefit from Width", ACCEPTED PAPER for the International Conference on Machine Learning  (ICML), 2025. 

• P. Cisneros-Velarde, “Bypassing Safety Guardrails in LLMs Using Humor," ACCEPTED PAPER for a workshop at the The 63rd Annual Meeting of the Association for Computational Linguistics (ACL), 2025.

• P. Cisneros-Velarde, “Large Language Models can Achieve Social Balance," 2024.

• A. Attali, P. Cisneros-Velarde, M Morales and N. Amato, “Discrete State-Action Abstraction via the Successor Representation," 2022.

Published:

• P. Cisneros-Velarde, "Biases in Opinion Dynamics in Multi-Agent Systems of Large Language Models: A Case Study on Funding Allocation," Findings of the Association for Computational Linguistics: NAACL 2025, 2025.

• P. Cisneros-Velarde, Z. Chen, S. Koyejo, A. Banerjee, "Optimization and Generalization Guarantees for Weight Normalization," Transactions on Machine Learning Research (TMLR), 2025.

• P. Cisneros-Velarde, Sanmi Koyejo, "Finite-sample Guarantees for Nash Q-learning with Linear Function Approximation," Uncertainty in Artificial Intelligence (UAI), 2023.

• A Banerjee, P. Cisneros-Velarde, L. Zhu and M Belkin, “Neural tangent kernel at initialization: linear width suffices,” Uncertainty in Artificial Intelligence (UAI), 2023.

• A. Banerjee, P. Cisneros-Velarde, L. Zhu, M. Belkin, “Restricted Strong Convexity of Deep Learning Models with Smooth Activations," International Conference on Learning Representations (ICLR), 2023.

• P. Cisneros-Velarde, B. Lyu, S. Koyejo and M. Kolar, “One Policy is Enough: Parallel Exploration with a Single Policy is Near-Optimal for Reward-Free Reinforcement Learning," International Conference on Artificial Intelligence and Statistics (AISTATS), 2023.

• P. Cisneros-Velarde and F. Bullo, “Distributed Wasserstein Barycenters via Displacement Interpolation,” IEEE Transactions on Control  of Network Systems,  2022.

• P. Cisneros-Velarde and F. Bullo, “A Contraction Theory Approach to Optimization Algorithms from Acceleration Flows,” International Conference on Artificial Intelligence and Statistics (AISTATS), 2022.

• Francesco Bullo, Pedro Cisneros-Velarde, Alexander Davydov, Saber Jafarpour, "From contraction theory to fixed point algorithms on Riemannian and non-Euclidean spaces," 60th IEEE Conference on Decision and Control (CDC), 2021.

• P. Cisneros-Velarde, S. Jafarpour and F. Bullo, “Contraction Theory for Dynamical Systems on Hilbert Spaces,” IEEE Transactions on Automatic Control, 2021.

• P. Cisneros-Velarde and F. Bullo, “Multi-group SIS Epidemics with Simplicial and Higher-Order Interactions,” IEEE Transactions on Control  of Network Systems, 2021.

• P. Cisneros-Velarde, S. Jafarpour and F. Bullo, “Distributed and time-varying primal-dual dynamics via contraction analysis,” IEEE Transactions on Automatic Control, 2021.

• P. Cisneros-Velarde and F. Bullo, “A Network Formation Game for the Emergence of Hierarchies,” PloS One, 2021.

• S. Jafarpour, P. Cisneros-Velarde and F. Bullo, “Weak and Semi-Contraction Theory with Application to Network Systems,” IEEE Transactions on Automatic Control, 2021.

• P. Cisneros-Velarde, K. S. Chan, and F. Bullo, “Polarization and Fluctuations in Signed Social Networks,” IEEE Transactions on Automatic Control, 2020.

• P. Cisneros-Velarde, N. E. Friedkin, A. V. Proskurnikov, F. Bullo, ”Structural Balance via Gradient Flows over Signed Graphs,” IEEE Transactions on Automatic Control, 2020.

• P. Cisneros-Velarde, S. Y. Oh, and A. Petersen, “Distributionally Robust Formulation and Model Selection for the Graphical Lasso,” International Conference on Artificial Intelligence and Statistics (AISTATS), 2020.

• P. Cisneros-Velarde and F. Bullo, ”Signed Network Formation Games and Clustering Balance,” Dynamic Games and Applications, 2020.

• W. Mei, P. Cisneros-Velarde, G. Chen, N. E. Friedkin and F. Bullo, ”Dynamic social balance and convergent appraisals via homophily and influence mechanisms,” Automatica, 2019.

• P. Cisneros-Velarde, D. Oliveira, and K. Chan, ”Spread and Control of Misinformation with Heterogeneous Agents,” International Workshop on Complex Networks, 2019.