ViTaX provides formal guarantees for targeted explanations in safety-critical systems.
problem Need trustworthy explanations for safety-critical deep neural networks.
method Formal reachability analysis for targeted, semifactual explanations.
result First method to provide formally guaranteed explanations of model resilience.
New IRL algorithm for continuous state spaces with formal guarantees.
problem Finding a reward function for expert behavior in continuous state spaces.
method Modeling the system using orthonormal functions and providing correctness proofs.
result Proof of correctness and formal guarantees on sample and time complexity.
Efficiently certifies global robustness of large neural networks with probabilistic guarantees.
problem Certifying robustness of large neural networks in a scalable and efficient manner.
method Sampling an ε-net and invoking a local robustness oracle.
result Certifies a probabilistic relaxation of robustness efficiently and globally.
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.
New method provides formal uncertainty guarantees for image classifiers.
problem Uncertainty quantification for image classifiers without formal guarantees.
method Adapts conformal prediction to give stable, formal coverage guarantees.
result Method outperforms existing approaches in coverage and set size.
We define the notion of a formal connection for a smooth family of star products with fixed underlying symplectic structure. Such a formal connection allows one to relate star products at different points in the family. This generalizes the formal Hitchin connection introduced by the first author. We establish a necess…
Formal methods verify continuous auctions at exchanges.
problem Ensuring fairness and correctness in continuous auctions.
method Formal specification, design, and verification of continuous double auctions.
result A verified algorithm satisfies natural properties of auctions.
KCRL learns stable policies for nonlinear systems with formal guarantees.
problem Lack of stabilization guarantees in RL methods for safety-critical systems.
method KCRL uses Krasovskii's Lyapunov functions as a stability constraint and a primal-dual approach to learn stabilizing policies.
result KCRL guarantees learning a stabilizing policy in a finite number of interactions.
LPF provides formal guarantees for aggregating multi-evidence in probabilistic tasks.
problem Lack of formal guarantees for multi-evidence reasoning in AI.
method LPF uses variational autoencoders and Sum-Product Networks to aggregate evidence items.
result Proves multiple formal guarantees including calibration preservation and error decay.
Method provides formal guarantees for decomposing model uncertainty.
problem Decomposing model uncertainty into aleatoric and epistemic components.
method Higher-order calibration using k-snapshots.
result Formal guarantees for aleatoric uncertainty matching real-world distribution.
Despite the functional success of deep neural networks (DNNs), their trustworthiness remains a crucial open challenge. To address this challenge, both testing and verification techniques have been proposed. But these existing techniques provide either scalability to large networks or formal guarantees, not both. In thi…
Paper formalizes multi-dimensional FSD using geometric methods.
problem Complex measure theory and calculus barriers to formalization in proof assistants.
method Geometric framework for first-order stochastic dominance in N dimensions.
result Geometric approach bypasses complex integration theory for direct comparison of survival probabilities.
New methods adapt conformal prediction to unknown subpopulation shifts.
problem Failure of conformal prediction under unknown subpopulation shifts.
method Proposes new methods that adapt conformal prediction to unknown subpopulation shifts without explicit subpopulation labels.
result Ensures valid coverage guarantees without explicit knowledge of subpopulation structure.
Meta-learning reduces set prediction size in conformal prediction for few-shot calibration.
problem Inefficient set prediction in conformal prediction for limited training data.
method Meta-learning approach using cross-validation-based conformal prediction.
result Meta-learning scheme reduces set prediction size and preserves formal guarantees.
New method provides scalable safety guarantees for RL agents.
problem Safe reinforcement learning in real-life scenarios.
method State-augmentation and shield design for probabilistic avoidance.
result Strict formal safety guarantees for RL agents, scalable and practical.
Recent work has shown that state-of-the-art classifiers are quite brittle, in the sense that a small adversarial change of an originally with high confidence correctly classified input leads to a wrong classification again with high confidence. This raises concerns that such classifiers are vulnerable to attacks and ca…
CLEVER (Cross-Lipschitz Extreme Value for nEtwork Robustness) is an Extreme Value Theory (EVT) based robustness score for large-scale deep neural networks (DNNs). In this paper, we propose two extensions on this robustness score. First, we provide a new formal robustness guarantee for classifier functions that are twic…
Bayesian inference and Gaussian processes are widely used in applications ranging from robotics and control to biological systems. Many of these applications are safety-critical and require a characterization of the uncertainty associated with the learning model and formal guarantees on its predictions. In this paper w…
New method removes specific training data influence from neural networks.
problem Removing specific training data influence from neural networks for privacy and regulatory reasons.
method Noisy fine-tuning on retain data to ensure provable unlearning guarantees without restrictive assumptions.
result Achieves formal unlearning guarantees and performs effectively in practice.
PITMonitor monitors model calibration over time with formal error guarantees.
problem Fixed-sample tests applied to models over time can lead to false alarms.
method PITMonitor uses mixture e-processes to detect distributional shifts in probability integral transforms.
result PITMonitor achieves competitive detection rates on river's FriedmanDrift benchmark.
Paper formalizes continual learning, proposing a feature extraction approach.
problem Learning from multiple environments without forgetting previous ones.
method Feature extraction framework and gradient-based algorithm DPGD.
result Efficient algorithm DPGD avoids catastrophic forgetting.
Learning disentangled representations that correspond to factors of variation in real-world data is critical to interpretable and human-controllable machine learning. Recently, concerns about the viability of learning disentangled representations in a purely unsupervised manner has spurred a shift toward the incorporat…
The formalism of anchor words has enabled the development of fast topic modeling algorithms with provable guarantees. In this paper, we introduce a protocol that allows users to interact with anchor words to build customized and interpretable topic models. Experimental evidence validating the usefulness of our approach…
DPNR preserves privacy of text representations using differential privacy.
problem Privacy leakage in deep learning text representations.
method DPNR uses Differential Privacy to provide formal privacy guarantees and dropout masking for enhanced privacy.
result DPNR reduces privacy leakage without significantly sacrificing main task performance.
Paper formalizes Simon's satisficing through FFSD, proving its equivalence to expected utility theory.
problem Formalizing Herbert Simon's bounded rationality concept in economic decision-making.
method Developed FFSD framework using Lean 4 theorem prover, proving equivalence to expected utility theory.
result Equivalence theorem linking FFSD to expected utility maximization for approximate indicator functions.
Paper formalizes analogy between data sets and models using Hoare logic.
problem Lack of formal criteria for transferring machine learning models between data domains.
method Formalization of analogy using first-order logic and Hoare logic, rigorous theorem proving.
result Rigorous formalization of analogy in knowledge transfer between machine learning models.
This work provides formal guarantees for heuristic optimization methods in machine learning.
problem Lack of theoretical understanding of heuristic optimization methods in machine learning.
method Analysis and formal guarantees for AdaGrad, SGD with exponential and cosine step sizes, and momentum methods.
result First formal guarantees for AdaGrad and SGD variants, including convergence and adaptivity to noise.
Simple proof shows forecasts can be calibrated in a few periods.
problem Ensuring forecasts are calibrated over multiple periods.
method Uses minimax theorem to prove existence and guarantees calibration error.
result Calibration can be achieved in N3 periods with error at most 1/N. Structured subsampling improves privacy in deep time series forecasting.
problem Incompatible privacy guarantees with time series forecasting.
method Structured subsampling of sequential data for privacy amplification.
result Structured subsampling enables training with strong privacy guarantees.
Paper formalizes and analyzes a new bound for variational inference.
problem Lack of theoretical guarantees in variational algorithms.
method Introduces VR-IWAE bound, a generalization of IWAE.
result VR-IWAE bound leads to unbiased gradient estimators.
Paper defines and solves a problem in representation learning to ensure fairness with high confidence.
problem Learning fair representations with high confidence guarantees for all downstream tasks.
method Formally defines the problem, introduces FRG framework, proves high probability fairness, and demonstrates effectiveness empirically.
result FRG framework provides high-confidence guarantees for limiting unfairness across all downstream models and tasks.
Geometric formalism views optimization algorithms as discrete connections, revealing their algebraic curvature and flatness properties.
problem Understanding and optimizing the behavior of iterative optimization algorithms.
method Introducing a geometric and operator-theoretic formalism where optimization algorithms are encoded by coupled channels (drift and diffusion) whose algebraic curvature measures the deviation from ideal reversibility.
result Flat connections correspond to methods whose updates commute up to higher order, achieving minimal numerical dissipation and preserving stability.
SAL framework uses unlabeled data to improve OOD detection.
problem Lack of clean OOD samples makes OOD detection challenging.
method SAL framework separates candidate outliers and trains an OOD classifier.
result SAL achieves state-of-the-art performance on benchmarks.
This paper introduces a new classification tool named Silas, which is built to provide a more transparent and dependable data analytics service. A focus of Silas is on providing a formal foundation of decision trees in order to support logical analysis and verification of learned prediction models. This paper describes…
New method protects whistleblowers from retaliation by ensuring their reports remain private.
problem Whistleblowers face retaliation, and current protections are insufficient.
method Formalizes protection against strong-adversary threat model as per-report (0,δ)-differential privacy, and provides a generic mechanism to reduce private auditing to private continual counting. result Demonstrates a reduction in selection error and improved utility over randomized response.
This paper formalizes reward training in RLHF and provides theoretical guarantees.
problem Costly human feedback collection in RLHF.
method Linear contextual dueling bandit method for regret minimization.
result Derives bounds on simple regret for offline reward training.
Fine-tuning neural networks to guarantee performance on specific examples can also introduce incorrect inputs.
problem Ensuring reliable performance of neural networks on specific examples.
method Using SMT solvers to fine-tune ReLU neural networks to guarantee outcomes on a finite set of particular examples.
result Fine-tuning can introduce incorrect inputs that trigger unexpected performance.
ConfHit provides valid guarantees for generative models without oracle access.
problem Reliable guarantees for novel candidate generation in generative models.
method Formalizes certification and refinement of generated sets, leveraging weighted exchangeability and density-ratio weighted conformal p-values.
result Consistently delivers valid coverage guarantees and compact certified sets across various generative tasks.
We present a new approach to assessing the robustness of neural networks based on estimating the proportion of inputs for which a property is violated. Specifically, we estimate the probability of the event that the property is violated under an input model. Our approach critically varies from the formal verification f…
This work improves deep reinforcement learning robustness to adversarial state uncertainty.
problem Robustness of deep reinforcement learning to adversarial state uncertainty.
method Certified adversarial robustness techniques are applied to deep reinforcement learning algorithms to compute guaranteed lower bounds on state-action values.
result The approach increases robustness to noise and adversaries in pedestrian collision avoidance and classic control tasks.
Develops statistical guarantees for image-to-image regression models.
problem Current image-to-image regression models lack statistical guarantees for model mistakes and hallucinations.
method Uncertainty quantification techniques with rigorous statistical guarantees for image-to-image regression problems.
result Derives uncertainty intervals around each pixel with formal mathematical guarantees.
This paper uses SLT to ensure learning guarantees in CD detection.
problem Lack of learning guarantees in CD detection algorithms.
method Adapting SLT assumptions to CD scenarios to ensure learning guarantees.
result Ensured learning guarantees in CD detection algorithms.
Neural networks are part of many contemporary NLP systems, yet their empirical successes come at the price of vulnerability to adversarial attacks. Previous work has used adversarial training and data augmentation to partially mitigate such brittleness, but these are unlikely to find worst-case adversaries due to the c…
Unified framework for subsampling mechanisms with tighter privacy guarantees.
problem Improving privacy in machine learning models through subsampling.
method Conditional optimal transport for deriving mechanism-specific subsampling guarantees.
result Tighter privacy bounds for subsampled mechanisms compared to traditional methods.
Privacy is enhanced by synthetic data release even with unlimited data.
problem Improving privacy guarantees for synthetic data release.
method Analyzing a bounded-parameter assumption to show privacy amplification persists with unlimited synthetic records.
result Privacy amplification is possible even with an unbounded number of synthetic records.
Statistical guarantees for hyperparameter selection
problem Hyperparameter selection in AI systems
method Learn-then-test framework
result Provable reliability and safety
Global-QSGD accelerates distributed training by up to 3.51%.
problem High communication overhead in distributed deep learning.
method Allreduce-compatible gradient quantization with theoretical guarantees.
result Global-QSGD accelerates distributed training by up to 3.51%.
We provide novel theoretical insights on structured prediction in the context of efficient convex surrogate loss minimization with consistency guarantees. For any task loss, we construct a convex surrogate that can be optimized via stochastic gradient descent and we prove tight bounds on the so-called "calibration func…