More Like this

1. The Capacity of 3 User Linear Computation Broadcast

   https://doi.org/10.1109/TIT.2024.3392685  

   Yao, Yuhang; Jafar, Syed A (June 2024, IEEE Transactions on Information Theory)

   The K User Linear Computation Broadcast (LCBC) problem is comprised of d dimensional data (from Fq), that is fully available to a central server, and K users, who require various linear computations of the data, and have prior knowledge of various linear functions of the data as side-information. The optimal broadcast cost is the minimum number of q-ary symbols to be broadcast by the server per computation instance, for every user to retrieve its desired computation. The reciprocal of the optimal broadcast cost is called the capacity. The main contribution of this paper is the exact capacity characterization for the K = 3 user LCBC for all cases, i.e., for arbitrary finite fields Fq, arbitrary data dimension d, and arbitrary linear side-informations and demands at each user. A remarkable aspect of the converse (impossibility result) is that unlike the 2 user LCBC whose capacity was determined previously, the entropic formulation (where the entropies of demands and side-informations are specified, but not their functional forms) is insufficient to obtain a tight converse for the 3 user LCBC. Instead, the converse exploits functional submodularity. Notable aspects of achievability include sufficiency of vector linear coding schemes, subspace decompositions that parallel those found previously by Yao Wang in degrees of freedom (DoF) studies of wireless broadcast networks, and efficiency tradeoffs that lead to a constrained waterfilling solution. Random coding arguments are invoked to resolve compatibility issues that arise as each user has a different view of the subspace decomposition, conditioned on its own side-information. 

   more » « less
2. Linear Stochastic Bandits over a Bit-Constrained Channel

   Mitra, Aritra; Hassani, Hamed; Pappas, George (June 2023, Learning for Dynamics and Control)

   One of the primary challenges in large-scale distributed learning stems from stringent communication constraints. While several recent works address this challenge for static optimization problems, sequential decision-making under uncertainty has remained much less explored in this regard. Motivated by this gap, we introduce a new linear stochastic bandit formulation over a bit-constrained channel. Specifically, in our setup, an agent interacting with an environment transmits encoded estimates of an unknown model parameter to a server over a communication channel of finite capacity. The goal of the server is to take actions based on these estimates to minimize cumulative regret. To this end, we develop a novel and general algorithmic framework that hinges on two main components: (i) an adaptive encoding mechanism that exploits statistical concentration bounds, and (ii) a decision-making principle based on confidence sets that account for encoding errors. As our main result, we prove that when the unknown model is d-dimensional, a channel capacity of O(d) bits suffices to achieve order-optimal regret. We also establish that for the simpler unstructured multi-armed bandit problem, 1 bit channel capacity is sufficient for achieving optimal regret bounds. 

   more » « less
3. Multiscale regression on unknown manifolds

   https://doi.org/10.3934/mine.2022028  

   Liao, Wenjing; Maggioni, Mauro; Vigogna, Stefano (January 2022, Mathematics in Engineering)

   null (Ed.)

   We consider the regression problem of estimating functions on $$ \mathbb{R}^D $$ but supported on a $ d $-dimensional manifold $$ \mathcal{M} ~~\subset \mathbb{R}^D $$ with $$ d \ll D $$. Drawing ideas from multi-resolution analysis and nonlinear approximation, we construct low-dimensional coordinates on $$ \mathcal{M} $$ at multiple scales, and perform multiscale regression by local polynomial fitting. We propose a data-driven wavelet thresholding scheme that automatically adapts to the unknown regularity of the function, allowing for efficient estimation of functions exhibiting nonuniform regularity at different locations and scales. We analyze the generalization error of our method by proving finite sample bounds in high probability on rich classes of priors. Our estimator attains optimal learning rates (up to logarithmic factors) as if the function was defined on a known Euclidean domain of dimension $ d $, instead of an unknown manifold embedded in $$ \mathbb{R}^D $$. The implemented algorithm has quasilinear complexity in the sample size, with constants linear in $ D $ and exponential in $ d $. Our work therefore establishes a new framework for regression on low-dimensional sets embedded in high dimensions, with fast implementation and strong theoretical guarantees. 

   more » « less
4. Linear Stochastic Bandits over a Bit-Constrained Channel

   Mitra, Aritra; Hassani, Hamed; Pappas, George (May 2023, Learning for Dynamics and Control Conference)

   One of the primary challenges in large-scale distributed learning stems from stringent communication constraints. While several recent works address this challenge for static optimization problems, sequential decision-making under uncertainty has remained much less explored in this regard. Motivated by this gap, we introduce a new linear stochastic bandit formulation over a bit-constrained channel. Specifically, in our setup, an agent interacting with an environment transmits encoded estimates of an unknown model parameter to a server over a communication channel of finite capacity. The goal of the server is to take actions based on these estimates to minimize cumulative regret. To this end, we develop a novel and general algorithmic framework that hinges on two main components: (i) an adaptive encoding mechanism that exploits statistical concentration bounds, and (ii) a decision-making principle based on confidence sets that account for encoding errors. As our main result, we prove that when the unknown model is d-dimensional, a channel capacity of O(d) bits suffices to achieve order-optimal regret. We also establish that for the simpler unstructured multi-armed bandit problem, 1 bit channel capacity is sufficient for achieving optimal regret bounds. Keywords: Linear Bandits, Distributed Learning, Communication Constraints 

   more » « less
5. Practical and Effective Heuristics for the Backhaul Profit Maximization Problem

   https://doi.org/10.1007/s11067-025-09684-0  

   Ryan, Daniel; Lam, Tran; Dong, Yuanyuan; Olinick, Eli (May 2025, Networks and Spatial Economics)

   Abstract Given an empty delivery vehicle, the backhaul profit maximization problem (BPMP) is to select a profit-maximizing subset of available pick-up-and-delivery requests to accept considering the vehicle’s capacity and a time limit for the vehicle to reach a specified destination or, equivalently, a driving-distance limit. Implemented in our computing environment, the fastest known exact algorithm for BPMP requires approximately 11 hours and 44 minutes on average to solve the largest instances in the literature, which have 70 to 80 potential pick-up/drop-off locations. The fastest available heuristic from the literature is considerably faster, and finds high quality solutions, but requires a state-of-the-art mixed-integer programming solver. We present a heuristic framework for the BPMP based on greedy construction, iterative local search, and randomization. Algorithms developed with the framework are implemented in the freely and widely available C++ language and their effectiveness is demonstrated through an extensive computational experiment on both benchmark and randomly generated problem instances. We find that our approach is competitive with approaches from the literature in solution quality as well as running time. 

   more » « less