Formalizes weak and strong verification for LLMs, controlling errors without assumptions.
problem Balancing cost and reliability in reasoning with LLMs.
method Formalizes weak-strong verification policies, introduces metrics, develops online algorithm.
result Optimal policies admit a two-threshold structure, and calibration and sharpness govern value of weak verifiers.
Revel tackles safe exploration in RL with verified symbolic policies.
problem Computational infeasibility of verifying neural networks in RL learning loops.
method Two policy classes: neurosymbolic with approximate gradients and symbolic policies for efficient verification. Mirror descent over policies to safely update and project policies.
result Revel discovers policies that outperform prior approaches to verified exploration.
New approach generates optimal disturbances for controller verification.
problem Optimizing disturbances for controller verification with blackbox access.
method Online learning approach that adaptively generates disturbances based on controller inputs.
result New algorithm (MOTR) outperforms existing methods in simulated examples.
SafePILCO is a Python tool for safe reinforcement learning.
problem Safe and efficient policy synthesis in reinforcement learning.
method Extends PILCO algorithm with safety features, implemented in Python.
result Safe and data-efficient policy synthesis achieved.
Deep neural networks are known to be fragile to small adversarial perturbations. This issue becomes more critical when a neural network is interconnected with a physical system in a closed loop. In this paper, we show how to combine recent works on neural network certification tools (which are mainly used in static set…
We present a reinforcement learning framework, called Programmatically Interpretable Reinforcement Learning (PIRL), that is designed to generate interpretable and verifiable agent policies. Unlike the popular Deep Reinforcement Learning (DRL) paradigm, which represents policies by neural networks, PIRL represents polic…
We study the problem of programmatic reinforcement learning, in which policies are represented as short programs in a symbolic language. Programmatic policies can be more interpretable, generalizable, and amenable to formal verification than neural policies; however, designing rigorous learning approaches for such poli…
The paper uses RL to verify hypotheses, overcoming existing limitations.
problem Verifying hypotheses using reinforcement learning.
method Formulated hypothesis verification as an RL problem, exploiting hypothesis structure.
result RL agents can successfully verify hypotheses, even those not factorizable.
Despite the tremendous advances that have been made in the last decade on developing useful machine-learning applications, their wider adoption has been hindered by the lack of strong assurance guarantees that can be made about their behavior. In this paper, we consider how formal verification techniques developed for …
Reinforcement learning is a powerful paradigm for learning optimal policies from experimental data. However, to find optimal policies, most reinforcement learning algorithms explore all possible actions, which may be harmful for real-world systems. As a consequence, learning algorithms are rarely applied on safety-crit…
Enhances large language models' reasoning through simpler off-policy reinforcement learning.
problem Improving large language models' ability to reason and solve problems.
method EM Policy Gradient, optimizing expected return over reasoning trajectories using Expectation-Maximization (EM) optimization.
result Achieves comparable or slightly superior performance to state-of-the-art methods on reasoning datasets, with additional cognitive behaviors.
VERAFI improves financial AI by verifying calculations and compliance.
problem Financial AI systems generate errors and violations during reasoning.
method VERAFI combines dense retrieval, reranking, and automated reasoning policies.
result VERAFI achieves 94.7% factual correctness, 81% relative improvement.
Develops a control framework for systemic risk under uncertainty.
problem Systemic risk under model uncertainty.
method Linear-quadratic mean-field control framework with viscosity solutions and verification theorems.
result Explicit feedback controls derived from a coupled Riccati system, preserving analytical tractability.
SIMPOL solves complex economic models using numerical methods.
problem Optimizing consumption and savings under uncertainty.
method SIMPOL uses a modular numerical framework combining policy iteration and finite difference schemes.
result SIMPOL produces solutions consistent with economic and mathematical theory.
Develops first robustness verification for complex Transformers.
problem Certify prediction behavior of Transformers with complex self-attention layers.
method Resolves challenges of cross-nonlinearity and cross-position dependency in Transformers.
result Certified robustness bounds are significantly tighter than Interval Bound Propagation.
In this paper, we assume an insure is allowed to purchase proportional reinsurance and can invest his or her wealth into the financial market where a savings account, stocks and bonds are available. Different from classical optimal investment and reinsurance problem, this paper studies the insurer's long-term investmen…
Paper develops PAC verification for hypothesis classes and statistical algorithms.
problem Verifying machine learning models interactively.
method Develops interactive proof for PAC verification, proves lower bounds, and introduces a generalization.
result Improved protocol for verifying unions of intervals and statistical query algorithms.
The trade-off between the cost of acquiring and processing data, and uncertainty due to a lack of data is fundamental in machine learning. A basic instance of this trade-off is the problem of deciding when to make noisy and costly observations of a discrete-time Gaussian random walk, so as to minimise the posterior var…
This paper examines the dividend and investment policies of a cash constrained firm that has access to costly external funding. We depart from the literature by allowing the firm to issue collateralized debt to increase its investment in productive assets resulting in a performance sensitive interest rate on debt. We f…
Hypothesis testing is an important problem with applications in target localization, clinical trials etc. Many active hypothesis testing strategies operate in two phases: an exploration phase and a verification phase. In the exploration phase, selection of experiments is such that a moderate level of confidence on the …
We explore the concept of co-design in the context of neural network verification. Specifically, we aim to train deep neural networks that not only are robust to adversarial perturbations but also whose robustness can be verified more easily. To this end, we identify two properties of network models - weight sparsity a…
Accelerates DNN robustness verification with target labels.
problem Improving the robustness of deep neural networks against adversarial attacks.
method Guiding robustness verification with target labels, reducing search space and using symbolic interval propagation and linear relaxation.
result Significantly improves DNN verification speed by 36X, especially when perturbation distance is reasonable.
New methods combat data poisoning attacks in bandit algorithms using limited verification.
problem Data poisoning attacks on bandit algorithms, especially in the UCB and ETC types.
method Verification-based mechanisms to restore optimal regret with limited verifications.
result A simple modified ETC type bandit algorithm can restore optimal regret with O(logT) verifications. Improved speaker verification with condition-aware backend.
problem Speaker verification calibration issues under unknown conditions.
method Discriminative PLDA model with joint training and condition integration.
result Out-of-the-box excellent calibration performance.
In this paper we demonstrate that performance of a speaker verification system can be improved by concatenating electroencephalography (EEG) signal features with speech signal features or only using EEG signal features. We use state-of-the-art end-to-end deep learning model for performing speaker verification and we de…
Improves neural network verification by merging abstract domains and Lagrangian methods.
problem Prove provable bounds for neural network outputs given input ranges.
method Uses zonotopes within a Lagrangian decomposition to verify deep neural networks.
result Yields bounds that improve upon existing techniques in both time and tightness.
In this article we extend earlier work on the jump-diffusion risk-sensitive asset management problem [SIAM J. Fin. Math. (2011) 22-54] by allowing jumps in both the factor process and the asset prices, as well as stochastic volatility and investment constraints. In this case, the HJB equation is a partial integro-diffe…
This paper explores formal verification for autonomous systems, identifying limitations and proposing improvements.
problem Ensuring safety of autonomous systems like self-driving cars and drones.
method Formal verification techniques based on formal methods, analyzing three assumptions and their limitations.
result Preliminary work to improve the strength of evidence provided by formal verification.
Paper develops a model for verifying facts in tables without pre-retrieved evidence.
problem Verification of factual claims in structured data, especially in open-domain settings.
method Joint reranking-and-verification model that fuses evidence documents.
result Model achieves comparable performance to closed-domain state-of-the-art on TabFact dataset.
We consider a model of optimal investment and consumption with both habit formation and partial observations in incomplete Itô processes market. The investor chooses his consumption under the addictive habits constraint while only observing the market stock prices but not the instantaneous rate of return. Applying the …
Model cash management under ambiguity using maxmin preferences and diffusion.
problem Optimizing cash reserves in the presence of ambiguity.
method Singular control model with maxmin preferences, verified using Dynkin games.
result Higher expected costs and narrower inaction region under increased ambiguity.
AgensFlow learns multi-agent coordination policies from experience.
problem Difficult coordination choices in multi-agent systems built on LLMs.
method Online policy learning from repeated trajectories, treating decisions as learnable.
result Learned routing improves coordination-heavy workflows over static wiring.
Fraud detection is a difficult problem that can benefit from predictive modeling. However, the verification of a prediction is challenging; for a single insurance policy, the model only provides a prediction score. We present a case study where we reflect on different instance-level model explanation techniques to aid …
Efficiently verifies neural networks by handling neuron splits, improving speed and accuracy.
problem Handling neuron split constraints in incomplete neural network verification.
method β-CROWN, which optimizes parameters β to encode neuron splits and uses them in bound propagation.
result β-CROWN significantly speeds up verification while maintaining high accuracy.
Semantify-NN verifies neural network robustness against semantic perturbations.
problem Verifying robustness of neural networks against semantic adversarial attacks.
method Inserting semantic perturbation layers (SP-layers) into neural networks to verify robustness.
result Semantify-NN significantly improves robustness verification performance over ℓp-norm-based methods. Researchers find floating point errors can mislead neural network verifiers.
problem Floating point arithmetic inaccuracies mislead neural network verifiers.
method Efficiently searches inputs and constructs neural network architectures to exploit verification errors.
result Floating point errors can systematically mislead neural network verifiers.
We study the robustness verification problem for tree-based models, including decision trees, random forests (RFs) and gradient boosted decision trees (GBDTs). Formal robustness verification of decision tree ensembles involves finding the exact minimal adversarial perturbation or a guaranteed lower bound of it. Existin…
We consider in this paper the optimal dividend problem for an insurance company whose uncontrolled reserve process evolves as a classical Cramér--Lundberg process. The firm has the option of investing part of the surplus in a Black--Scholes financial market. The objective is to find a strategy consisting of both invest…
This study extends verifiable learning to boosted tree ensembles, enabling efficient security verification.
problem Efficiently verifying the robustness of boosted tree ensembles against norm-based attackers.
method Formal verification of robustness for large-spread boosted tree ensembles, considering L∞-norm and pseudo-polynomial time for Lp-norm verification. result Polynomial time verification for L∞-norm attackers, NP-hard for other norms, and pseudo-polynomial time for Lp-norm verification. Accelerating Speculative Diffusions via Block Verification
problem Adapting speculative decoding for continuous diffusion models
method Introducing a novel speculative sampling mechanism for diffusion models
result Improves acceptance rate and speeds up inference
This paper tackles robustness of ensemble stumps and trees under general ℓ_p norm perturbations.
problem The vulnerability of ensemble stumps and trees to small input perturbations under the ℓ_∞ norm.
method Developed dynamic programming algorithms for robustness verification and certified defense under general ℓ_p norm perturbations.
result First certified defense method for ensemble stumps and trees under ℓ_p norm perturbations.
A new method for verifying deep learning architectures on FPGAs is proposed.
problem Design-time verification of deep learning architectures on FPGAs.
method 2-Level 3-Way (2L-3W) hardware-software co-verification methodology.
result Layer-by-layer similarity scores of 99% accuracy for successful mappings.
A neural network learns to improve verification of neural networks.
problem Efficient verification of neural networks for safety-critical applications.
method A graph neural network (GNN) learns to imitate strong branching heuristics for effective branching in the Branch and Bound (BaB) formulation.
result Reduces the number of branches and verification time by roughly 50% compared to hand-designed strategies.
GPUPoly verifies large neural networks robustly on GPUs.
problem Proving robustness of deep neural networks is crucial but challenging.
method Custom polyhedra algorithms on GPUs.
result GPUPoly can verify 1M neuron networks in 34.5 ms.
New spoofing strategies show PoL verification is more vulnerable than previously thought.
problem Vulnerability of Proof-of-Learning verification mechanisms.
method Developed new spoofing strategies that can be reproduced across different configurations and are more cost-effective.
result Current PoL verification is not robust to adversaries and requires further understanding of optimization in deep learning.
NeuroDiff improves neural network equivalence verification with fine-grained approximations.
problem Verifying the equivalence of compressed neural networks.
method Symbolic and fine-grained approximation technique for differential verification.
result NeuroDiff achieves up to 1000X speedup and 5X accuracy improvement.
Novel approach using Lagrangian Decomposition for neural network verification.
problem Computing tight bounds on neural network outputs.
method Lagrangian Decomposition, efficient supergradient ascent, proximal algorithm.
result Proves tighter bounds than previous dual algorithms.
Paper improves speaker verification with federated learning and differential privacy.
problem Improving speaker verification accuracy using private data.
method Combining federated learning and differential privacy to train an auxiliary model that predicts vocal characteristics.
result 6% relative improvement in equal error rate over a baseline system.