About

Pramod Viswanath is the Forrest G. Hamrick Professor in Engineering at Princeton University's Department of Electrical and Computer Engineering. His research interests include the principle understanding and design of blockchains, as well as inventing communication algorithms via deep learning and full stack design of blockchain technologies. He has contributed to the development of various blockchain protocols and algorithms, including consensus algorithms, cryptographic hash aggregators, off-chain networking stacks, and incentive structures for proof of stake blockchains. Additionally, he has extensive experience in wireless communications, having worked actively in the field for over a decade, focusing on modeling, algorithm design, and system solutions, including early work on the first OFDM-based cellular system and coauthoring a popular book on wireless communication. His work emphasizes theoretical and conceptual aspects of communication and network systems.

Research topics

• Computer science
• Computer network
• Mathematics
• Algorithm
• Theoretical computer science

Selected publications

• TabularMath: Evaluating Computational Extrapolation in Tabular Learning via Program-Verified Synthesis

  arXiv (Cornell University) · 2026-01-25

  articleOpen access

  Standard tabular benchmarks mainly focus on the evaluation of a model's capability to interpolate values inside a data manifold, where models good at performing local statistical smoothing are rewarded. However, there exists a very large category of high-value tabular data, including financial modeling and physical simulations, which are generated based upon deterministic computational processes, as opposed to stochastic and noisy relationships. Therefore, we investigate if tabular models can provide an extension from statistical interpolation to computational extrapolation. We propose TabularMath, a diagnostic benchmark of 114 deterministic problems (233,472 rows) generated from verified programs based on GSM8K and AIME. We evaluate 9 tabular architectures and in-context learning (ICL) with GPT-OSS-120B. On standard regression metrics, TabPFN v2.5 performs remarkably well, achieving R^2=0.998 in-distribution and maintaining positive R^2 even under distribution shift, which is unique among the tabular models we tested. When we measure rounded consistency (exact integer match), a different picture emerges: TabPFN v2.5 drops below 10% on out-of-distribution data, while ICL maintains around 40%. This gap between R^2 and exact-match accuracy suggests that tabular models learn smooth function approximations but struggle to recover precise computational outputs under extrapolation. The two paradigms appear complementary: TabPFN scales efficiently with data; ICL achieves exact computation from few examples. We release all code and data to support further investigation.

  PublisherOA PDF

• Correct Answers from Sound Reasoning: Verifiable Process Supervision for Language Models

  arXiv (Cornell University) · 2026-04-03

  preprintOpen access

  Training language models to produce both correct answers and sound reasoning remains an open challenge. Reinforcement learning with verifiable rewards typically optimizes only final outcomes, which can lead to a failure mode where task accuracy improves while reasoning becomes less accurate, less complete, or even internally inconsistent. We propose verifiable process supervision (VPS), a post-training framework for verifiable domains that jointly optimizes prediction accuracy and reasoning quality. We first apply supervised fine-tuning to induce a structured reasoning format, enabling syntactic extraction of intermediate claims that are evaluated against ground-truth signals to form process-level rewards. To address the heterogeneous difficulty of reasoning subtasks, we introduce adaptive reward weighting that prioritizes components with the largest remaining errors, creating an implicit curriculum. We evaluate VPS on chess, a controlled testbed where reasoning steps can be deterministically verified against engine signals. While accuracy-only RL improves move accuracy, it sharply degrades reasoning quality, increasing win-rate error by up to 112% and reducing internal consistency by up to 69%. In contrast, VPS preserves accuracy while significantly improving reasoning quality, reducing win-rate error by up to 30% and restoring consistency to near saturation. At matched accuracy, judge evaluation also prefers the process-supervised models. A reasoning-space analysis further shows that, without a structured prior, accuracy-only RL converges to budget-dependent shortcuts rather than sound multi-step reasoning. These results show that VPS enables language models to reason both accurately and reliably in verifiable domains.

  PublisherDOI

• ParlayMarket: Automated Market Making for Parlay-style Joint Contracts

  arXiv (Cornell University) · 2026-03-23

  preprintOpen accessSenior author

  Prediction markets are powerful mechanisms for information aggregation, but existing designs are optimized for single-event contracts. In practice, traders frequently express beliefs about joint outcomes - through parlays in sports, conditional forecasts across related events, or scenario bets in financial markets. Current platforms either prohibit such trades or rely on ad hoc mechanisms that ignore correlation structure, resulting in inefficient prices and fragmented liquidity. We introduce ParlayMarket, the first automated market-making design that supports parlay-style joint contracts within a unified liquidity pool while maintaining coherent pricing across base markets and their combinations. Our main result is a convergence characterization of the resulting system. Under repeated trading, the AMM dynamics converge to a unique fixed point corresponding to the best approximation to the true joint distribution within the model class. We show that (i) parameter error remains bounded at stationarity due to a balance between signal and noise in trade-induced updates, and (ii) pricing error and monetary loss scale with this parameter error, implying that aggregate market-maker loss remains controlled and grows at most quadratically in the number of base markets. These results establish explicit limits on the information-retrieval error achievable through the trading interface. Importantly, parlay trades play a structural role in this convergence: by providing direct constraints on joint outcomes, they improve identifiability of dependence structure and reduce steady-state error relative to markets that rely only on marginal trades. Empirically, we show both in controlled simulations and in replay on historical Kalshi parlay data that this design achieves the intended scaling while remaining effective in realistic market settings.

  PublisherDOI

• TabularMath: Evaluating Computational Extrapolation in Tabular Learning via Program-Verified Synthesis

  Open MIND · 2026-01-25

  preprint

  Standard tabular benchmarks mainly focus on the evaluation of a model's capability to interpolate values inside a data manifold, where models good at performing local statistical smoothing are rewarded. However, there exists a very large category of high-value tabular data, including financial modeling and physical simulations, which are generated based upon deterministic computational processes, as opposed to stochastic and noisy relationships. Therefore, we investigate if tabular models can provide an extension from statistical interpolation to computational extrapolation. We propose TabularMath, a diagnostic benchmark of 114 deterministic problems (233,472 rows) generated from verified programs based on GSM8K and AIME. We evaluate 9 tabular architectures and in-context learning (ICL) with GPT-OSS-120B. On standard regression metrics, TabPFN v2.5 performs remarkably well, achieving R^2=0.998 in-distribution and maintaining positive R^2 even under distribution shift, which is unique among the tabular models we tested. When we measure rounded consistency (exact integer match), a different picture emerges: TabPFN v2.5 drops below 10% on out-of-distribution data, while ICL maintains around 40%. This gap between R^2 and exact-match accuracy suggests that tabular models learn smooth function approximations but struggle to recover precise computational outputs under extrapolation. The two paradigms appear complementary: TabPFN scales efficiently with data; ICL achieves exact computation from few examples. We release all code and data to support further investigation.

  DOI

• ParlayMarket: Automated Market Making for Parlay-style Joint Contracts

  ArXiv.org · 2026-03-23

  articleOpen accessSenior author

  Prediction markets are powerful mechanisms for information aggregation, but existing designs are optimized for single-event contracts. In practice, traders frequently express beliefs about joint outcomes - through parlays in sports, conditional forecasts across related events, or scenario bets in financial markets. Current platforms either prohibit such trades or rely on ad hoc mechanisms that ignore correlation structure, resulting in inefficient prices and fragmented liquidity. We introduce ParlayMarket, the first automated market-making design that supports parlay-style joint contracts within a unified liquidity pool while maintaining coherent pricing across base markets and their combinations. Our main result is a convergence characterization of the resulting system. Under repeated trading, the AMM dynamics converge to a unique fixed point corresponding to the best approximation to the true joint distribution within the model class. We show that (i) parameter error remains bounded at stationarity due to a balance between signal and noise in trade-induced updates, and (ii) pricing error and monetary loss scale with this parameter error, implying that aggregate market-maker loss remains controlled and grows at most quadratically in the number of base markets. These results establish explicit limits on the information-retrieval error achievable through the trading interface. Importantly, parlay trades play a structural role in this convergence: by providing direct constraints on joint outcomes, they improve identifiability of dependence structure and reduce steady-state error relative to markets that rely only on marginal trades. Empirically, we show both in controlled simulations and in replay on historical Kalshi parlay data that this design achieves the intended scaling while remaining effective in realistic market settings.

  PublisherOA PDF

• VeRA: Verified Reasoning Data Augmentation at Scale

  ArXiv.org · 2026-01-23

  articleOpen access

  The main issue with most evaluation schemes today is their "static" nature: the same problems are reused repeatedly, allowing for memorization, format exploitation, and eventual saturation. To measure genuine AI progress, we need evaluation that is robust by construction, not by post-hoc detection. In response, we propose VeRA (Verified Reasoning Data Augmentation), a framework that converts benchmark problems into executable specifications, comprising (i) a natural language template with placeholder slots, (ii) a coherent generator that samples valid configurations, and (iii) a deterministic verifier that validates parameters and calculates the corresponding correct answers for each configuration. From a single seed problem, VeRA automatically creates unlimited verified variants with reliable labels at near-zero marginal cost without human involvement. VeRA operates in two complementary modes. VeRA-E (equivalent) rewrites problems while keeping the underlying logic intact, useful for detecting memorization versus genuine reasoning. VeRA-H (hardened) systematically increases complexity while remaining verifiable, enabling reliable creation and labelling of fresh difficult tasks at the boundary of intelligence. Evaluating 16 frontier models with VeRA, we find: (i) VeRA-E improves evaluation quality and reveals contamination patterns. (ii) VeRA-H enables human-free generation of hard tasks with reliable labels. (iii) VeRA establishes verified benchmarks as a general paradigm. VeRA reconceptualizes benchmarks from static objects used until exhausted, to executable specifications generating fresh, verified instances on demand, enhancing robustness and cost-effectiveness for evaluation. With VeRA, we envision that evaluation in any verifiable domain can scale indefinitely without sacrificing label integrity. To stimulate future research, we have open-sourced all code and datasets.

  PublisherOA PDF

• MEMO: Memory-Augmented Model Context Optimization for Robust Multi-Turn Multi-Agent LLM Games

  arXiv (Cornell University) · 2026-03-09

  preprintOpen access

  Multi-turn, multi-agent LLM game evaluations often exhibit substantial run-to-run variance. In long-horizon interactions, small early deviations compound across turns and are amplified by multi-agent coupling. This biases win rate estimates and makes rankings unreliable across repeated tournaments. Prompt choice worsens this further by producing different effective policies. We address both instability and underperformance with MEMO (Memory-augmented MOdel context optimization), a self-play framework that optimizes inference-time context by coupling retention and exploration. Retention maintains a persistent memory bank that stores structured insights from self-play trajectories and injects them as priors during later play. Exploration runs tournament-style prompt evolution with uncertainty-aware selection via TrueSkill, and uses prioritized replay to revisit rare and decisive states. Across five text-based games, MEMO raises mean win rate from 25.1% to 49.5% for GPT-4o-mini and from 20.9% to 44.3% for Qwen-2.5-7B-Instruct, using $2,000$ self-play games per task. Run-to-run variance also drops, giving more stable rankings across prompt variations. These results suggest that multi-agent LLM game performance and robustness have substantial room for improvement through context optimization. MEMO achieves the largest gains in negotiation and imperfect-information games, while RL remains more effective in perfect-information settings. All code is open-source and available here: https://github.com/openverse-ai/MEMO

  PublisherDOI

• Correct Answers from Sound Reasoning: Verifiable Process Supervision for Language Models

  ArXiv.org · 2026-04-03

  articleOpen access

  Training language models to produce both correct answers and sound reasoning remains an open challenge. Reinforcement learning with verifiable rewards typically optimizes only final outcomes, which can lead to a failure mode where task accuracy improves while reasoning becomes less accurate, less complete, or even internally inconsistent. We propose verifiable process supervision (VPS), a post-training framework for verifiable domains that jointly optimizes prediction accuracy and reasoning quality. We first apply supervised fine-tuning to induce a structured reasoning format, enabling syntactic extraction of intermediate claims that are evaluated against ground-truth signals to form process-level rewards. To address the heterogeneous difficulty of reasoning subtasks, we introduce adaptive reward weighting that prioritizes components with the largest remaining errors, creating an implicit curriculum. We evaluate VPS on chess, a controlled testbed where reasoning steps can be deterministically verified against engine signals. While accuracy-only RL improves move accuracy, it sharply degrades reasoning quality, increasing win-rate error by up to 112% and reducing internal consistency by up to 69%. In contrast, VPS preserves accuracy while significantly improving reasoning quality, reducing win-rate error by up to 30% and restoring consistency to near saturation. At matched accuracy, judge evaluation also prefers the process-supervised models. A reasoning-space analysis further shows that, without a structured prior, accuracy-only RL converges to budget-dependent shortcuts rather than sound multi-step reasoning. These results show that VPS enables language models to reason both accurately and reliably in verifiable domains.

  PublisherOA PDF

• VeRA: Verified Reasoning Data Augmentation at Scale

  Open MIND · 2026-01-23

  preprint

  The main issue with most evaluation schemes today is their "static" nature: the same problems are reused repeatedly, allowing for memorization, format exploitation, and eventual saturation. To measure genuine AI progress, we need evaluation that is robust by construction, not by post-hoc detection. In response, we propose VeRA (Verified Reasoning Data Augmentation), a framework that converts benchmark problems into executable specifications, comprising (i) a natural language template with placeholder slots, (ii) a coherent generator that samples valid configurations, and (iii) a deterministic verifier that validates parameters and calculates the corresponding correct answers for each configuration. From a single seed problem, VeRA automatically creates unlimited verified variants with reliable labels at near-zero marginal cost without human involvement. VeRA operates in two complementary modes. VeRA-E (equivalent) rewrites problems while keeping the underlying logic intact, useful for detecting memorization versus genuine reasoning. VeRA-H (hardened) systematically increases complexity while remaining verifiable, enabling reliable creation and labelling of fresh difficult tasks at the boundary of intelligence. Evaluating 16 frontier models with VeRA, we find: (i) VeRA-E improves evaluation quality and reveals contamination patterns. (ii) VeRA-H enables human-free generation of hard tasks with reliable labels. (iii) VeRA establishes verified benchmarks as a general paradigm. VeRA reconceptualizes benchmarks from static objects used until exhausted, to executable specifications generating fresh, verified instances on demand, enhancing robustness and cost-effectiveness for evaluation. With VeRA, we envision that evaluation in any verifiable domain can scale indefinitely without sacrificing label integrity. To stimulate future research, we have open-sourced all code and datasets.

  DOI

• MEMO: Memory-Augmented Model Context Optimization for Robust Multi-Turn Multi-Agent LLM Games

  ArXiv.org · 2026-03-09

  articleOpen access

  Multi-turn, multi-agent LLM game evaluations often exhibit substantial run-to-run variance. In long-horizon interactions, small early deviations compound across turns and are amplified by multi-agent coupling. This biases win rate estimates and makes rankings unreliable across repeated tournaments. Prompt choice worsens this further by producing different effective policies. We address both instability and underperformance with MEMO (Memory-augmented MOdel context optimization), a self-play framework that optimizes inference-time context by coupling retention and exploration. Retention maintains a persistent memory bank that stores structured insights from self-play trajectories and injects them as priors during later play. Exploration runs tournament-style prompt evolution with uncertainty-aware selection via TrueSkill, and uses prioritized replay to revisit rare and decisive states. Across five text-based games, MEMO raises mean win rate from 25.1% to 49.5% for GPT-4o-mini and from 20.9% to 44.3% for Qwen-2.5-7B-Instruct, using $2,000$ self-play games per task. Run-to-run variance also drops, giving more stable rankings across prompt variations. These results suggest that multi-agent LLM game performance and robustness have substantial room for improvement through context optimization. MEMO achieves the largest gains in negotiation and imperfect-information games, while RL remains more effective in perfect-information settings. All code is open-source and available here: https://github.com/openverse-ai/MEMO

  PublisherOA PDF

Recent grants

• CIF: Small: Cooperative Interference Management-A Fundamental Study

  NSF · $443k · 2010–2014

• CAREER: Opportunistic Communication: A Design Paradigm for Wireless Systems

  NSF · $400k · 2003–2009

• NeTS-WN: Collaborative Research: Interference Management and Cooperation in Wireless Networks: A Modern View

  NSF · $250k · 2007–2011

• Collaborative Research: MLWiNS:Physical Layer Communication revisited via Deep Learning

  NSF · $222k · 2020–2022

• Collaborative Research: CIF: Small: Designing Plotkin Transform Codes via Machine Learning

  NSF · $300k · 2023–2027

Frequent coauthors

• Sreeram Kannan

  101 shared

• Sewoong Oh

  Google (United States)

  88 shared

• David Tse

  45 shared

• Giulia Fanti

  29 shared

• Shaileshh Bojja Venkatakrishnan

  29 shared

• Yihan Jiang

  University of Florida

  24 shared

• Peter Kairouz

  24 shared

• Adnan Raja

  24 shared

Labs

• Pramod Viswanath LabPI

Education

• Doctor of Philosophy, Electrical Engineering and Computer Science

  University of California, Berkeley

  2000

• Master of Engineering

  Indian Institute of Science Bangalore

  1995

• Bachelor of Engineering

  National Institute of Technology Karnataka

  1993

Awards & honors

• Best Paper Award, Sigmetrics conference, 2015
• Xerox Faculty Research Award, College of Engineering, UIUC,…
• NSF CAREER Award, 2002
• Eliahu Jury Award, UC Berkeley, EECS, 2000
• Bernard Friedman Prize, UC Berkeley, Mathematics, 2000