Machine learning classification is used in numerous settings nowadays, such as medical or genomics predictions, spam detection, face recognition, and financial predictions.Due to privacy concerns, some of these applications, it important that the data classifier remain confidential.In this work, we construct three major protocols satisfy constraint: hyperplane decision, Naïve Bayes, decision trees.We also enable be combined with AdaBoost.At basis constructions a new library building blocks...

Silo is a new in-memory database that achieves excellent performance and scalability on modern multicore machines. was designed from the ground up to use system memory caches efficiently. For instance, it avoids all centralized contention points, including of transaction ID assignment. Silo's key contribution commit protocol based optimistic concurrency control provides serializability while avoiding shared-memory writes for records were only read. Though this might seem complicate...

MONOMI is a system for securely executing analytical workloads over sensitive data on an untrusted database server. works by encrypting the entire and running queries encrypted data. introduces split client/server query execution, which can execute arbitrarily complex data, as well several techniques that improve performance such workloads, including per-row precomputation, space-efficient encryption, grouped homomorphic addition, pre-filtering. Since these optimizations are good some but...

We consider adaptive control of the Linear Quadratic Regulator (LQR), where an unknown linear system is controlled subject to quadratic costs. Leveraging recent developments in estimation systems and robust controller synthesis, we present first provably polynomial time algorithm that provides high probability guarantees sub-linear regret on this problem. further study interplay between minimization parameter by proving a lower bound expected terms exploration schedule used any algorithm....

In this paper we study the problem of recovering a low-rank matrix from linear measurements. Our algorithm, which call Procrustes Flow, starts an initial estimate obtained by thresholding scheme followed gradient descent on non-convex objective. We show that as long measurements obey standard restricted isometry property, our algorithm converges to unknown at geometric rate. case Gaussian measurements, such convergence occurs for $n_1 \times n_2$ rank $r$ when number exceeds constant times...

We prove that the ordinary least-squares (OLS) estimator attains nearly minimax optimal performance for identification of linear dynamical systems from a single observed trajectory. Our upper bound relies on generalization Mendelson's small-ball method to dependent data, eschewing use standard mixing-time arguments. lower bounds reveal these match up logarithmic factors. In particular, we capture correct signal-to-noise behavior problem, showing more unstable are easier estimate. This is...

This paper addresses the optimal control problem known as Linear Quadratic Regulator in case when dynamics are unknown. We propose a multi-stage procedure, called Coarse-ID control, that estimates model from few experimental trials, error with respect to truth, and then designs controller using both uncertainty estimate. Our technique uses contemporary tools random matrix theory bound estimation procedure. also employ recently developed approach synthesis System Level Synthesis enables...

The traditional wisdom for building disk-based relational database management systems (DBMS) is to organize data in heavily-encoded blocks stored on disk, with a main memory block cache. In order improve performance given high disk latency, these use multi-threaded architecture dynamic record-level locking that allows multiple transactions access the at same time. Previous research has shown this results substantial overhead on-line transaction processing (OLTP) applications [15]. next...

We study the constrained linear quadratic regulator with unknown dynamics, addressing tension between safety and exploration in data-driven control techniques. present a framework which allows for system identification through persistent excitation, while maintaining by guaranteeing satisfaction of state input constraints. This involves novel method synthesizing robust constraint-satisfying feedback controllers, leveraging newly developed tools from level synthesis. connect statistical...

Inspired by the success of imitation and inverse reinforcement learning in replicating expert behavior through optimal control, we propose a based approach to safe controller synthesis on control barrier functions (CBFs). We consider setting known nonlinear affine dynamical system assume that have access trajectories generated an - practical example such would be kinematic model self-driving vehicle with (e.g., avoid collisions obstacles environment) human driver. then analyze optimization...

Multicore in-memory databases for modern machines can support extraordinarily high transaction rates online processing workloads. A potential weakness, however, is recovery from crash failures. Can classical techniques, such as checkpoints, be made both efficient enough to keep up with current systems' memory sizes and rates, smart avoid additional contention? Starting an multicore database system, we show that naive logging checkpoints make normal-case execution slower, but frequent disk...

Large language models (LLMs) exhibit a wide range of promising capabilities -- from step-by-step planning to commonsense reasoning that may provide utility for robots, but remain prone confidently hallucinated predictions. In this work, we present KnowNo, which is framework measuring and aligning the uncertainty LLM-based planners such they know when don't ask help needed. KnowNo builds on theory conformal prediction statistical guarantees task completion while minimizing human in complex...

Machine and reinforcement learning (RL) are increasingly being applied to plan control the behavior of autonomous systems interacting with physical world. Examples include self-driving vehicles, distributed sensor networks, agile robots. However, when machine is be in these new settings, algorithms had better come same type reliability, robustness, safety bounds that hallmarks theory, or failures could catastrophic. Thus, as more aggressively deployed critical it imperative theorists join...

This work explores the trade-off between number of samples required to accurately build models dynamical systems and degradation performance in various control objectives due a coarse approximation. In particular, we show that simple can be easily fit from input/output data are sufficient for achieving objectives. We derive bounds on noisy stable linear time-invariant system guarantee corresponding finite impulse response approximation is close true $\mathcal{H}_\infty$-norm. demonstrate...

Reinforcement learning (RL) has been successfully used to solve many continuous control tasks. Despite its impressive results however, fundamental questions regarding the sample complexity of RL on problems remain open. We study performance in this setting by considering behavior Least-Squares Temporal Difference (LSTD) estimator classic Linear Quadratic Regulator (LQR) problem from optimal control. give first finite-time analysis number samples needed estimate value function for a fixed...

We study the performance of certainty equivalent controller on Linear Quadratic (LQ) control problems with unknown transition dynamics. show that for both fully and partially observed settings, sub-optimality gap between cost incurred by playing true system using optimal LQ enjoys a fast statistical rate, scaling as square parameter error. To best our knowledge, result is first guarantee in Gaussian (LQG) setting. Furthermore, Regulator (LQR), improves upon recent work Dean et al. (2017),...

Many existing tools in nonlinear control theory for establishing stability or safety of a dynamical system can be distilled to the construction certificate function that guarantees desired property. However, algorithms synthesizing functions typically require closed-form analytical expression underlying dynamics, which rules out their use on many modern robotic platforms. To circumvent this issue, we develop learning only from trajectory data. We establish bounds generalization error -...

The need for robust control laws is especially important in safety-critical applications. We propose hybrid barrier functions as a means to synthesize that ensure safety. Based on this notion, we formulate an optimization problem learning from data. identify sufficient conditions the data such feasibility of ensures correctness learned functions. Our techniques allow us safely expand region attraction compass gait walker subject model uncertainty.