Searcharxiv⌕ Search

arXiv subjects

Suman Banerjee

Publications and source records attributed to Suman Banerjee.

At least 73 records · Page 4Linked to original sources
arXiv2024

A Picture is Worth 500 Labels: A Case Study of Demographic Disparities in Local Machine Learning Models for Instagram and TikTok

Mobile apps have embraced user privacy by moving their data processing to the user's smartphone. Advanced machine learning (ML) models, such as vision models, can now locally analyze user images to extract insights that drive several functionalities. Capitalizing on this new processing model of locally analyzing user images, we analyze two popular social media apps, TikTok and Instagram, to reveal (1) what insights vision models in both apps infer about users from their image and video data and (2) whether these models exhibit performance disparities with respect to demographics. As vision models provide signals for sensitive technologies like age verification and facial recognition, understanding potential biases in these models is crucial for ensuring that users receive equitable and accurate services. We develop a novel method for capturing and evaluating ML tasks in mobile apps, overcoming challenges like code obfuscation, native code execution, and scalability. Our method comprises ML task detection, ML pipeline reconstruction, and ML performance assessment, specifically focusing on demographic disparities. We apply our methodology to TikTok and Instagram, revealing significant insights. For TikTok, we find issues in age and gender prediction accuracy, particularly for minors and Black individuals. In Instagram, our analysis uncovers demographic disparities in the extraction of over 500 visual concepts from images, with evidence of spurious correlations between demographic features and certain concepts.

cs.LG↗
arXiv2024

Exploring the Design Space of Optical See-through AR Head-Mounted Displays to Support First Responders in the Field

First responders (FRs) navigate hazardous, unfamiliar environments in the field (e.g., mass-casualty incidents), making life-changing decisions in a split second. AR head-mounted displays (HMDs) have shown promise in supporting them due to its capability of recognizing and augmenting the challenging environments in a hands-free manner. However, the design space have not been thoroughly explored by involving various FRs who serve different roles (e.g., firefighters, law enforcement) but collaborate closely in the field. We interviewed 26 first responders in the field who experienced a state-of-the-art optical-see-through AR HMD, as well as its interaction techniques and four types of AR cues (i.e., overview cues, directional cues, highlighting cues, and labeling cues), soliciting their first-hand experiences, design ideas, and concerns. Our study revealed both generic and role-specific preferences and needs for AR hardware, interactions, and feedback, as well as identifying desired AR designs tailored to urgent, risky scenarios (e.g., affordance augmentation to facilitate fast and safe action). While acknowledging the value of AR HMDs, concerns were also raised around trust, privacy, and proper integration with other equipment. Finally, we derived comprehensive and actionable design guidelines to inform future AR systems for in-field FRs.

cs.HC↗
arXiv2024

Minimizing Regret in Billboard Advertisement under Zonal Influence Constraint

In a typical billboard advertisement technique, a number of digital billboards are owned by an influence provider, and many advertisers approach the influence provider for a specific number of views of their advertisement content on a payment basis. If the influence provider provides the demanded or more influence, then he will receive the full payment or else a partial payment. In the context of an influence provider, if he provides more or less than an advertiser's demanded influence, it is a loss for him. This is formalized as 'Regret', and naturally, in the context of the influence provider, the goal will be to allocate the billboard slots among the advertisers such that the total regret is minimized. In this paper, we study this problem as a discrete optimization problem and propose four solution approaches. The first one selects the billboard slots from the available ones in an incremental greedy manner, and we call this method the Budget Effective Greedy approach. In the second one, we introduce randomness with the first one, where we perform the marginal gain computation for a sample of randomly chosen billboard slots. The remaining two approaches are further improvements over the second one. We analyze all the algorithms to understand their time and space complexity. We implement them with real-life trajectory and billboard datasets and conduct a number of experiments. It has been observed that the randomized budget effective greedy approach takes reasonable computational time while minimizing the regret.

cs.DB↗
arXiv2024

Towards Regret Free Slot Allocation in Billboard Advertisement

Creating and maximizing influence among the customers is one of the central goals of an advertiser, and hence, remains an active area of research in recent times. In this advertisement technique, the advertisers approach an influence provider for a specific number of views of their content on a payment basis. Now, if the influence provider can provide the required number of views or more, he will receive the full, else a partial payment. In the context of an influence provider, it is a loss for him if he offers more or less views. This is formalized as 'Regret', and naturally, in the context of the influence provider, the goal will be to minimize this quantity. In this paper, we solve this problem in the context of billboard advertisement and pose it as a discrete optimization problem. We propose four efficient solution approaches for this problem and analyze them to understand their time and space complexity. We implement all the solution methodologies with real-life datasets and compare the obtained results with the existing solution approaches from the literature. We observe that the proposed solutions lead to less regret while taking less computational time.

cs.IR↗
arXiv2023

One continuous parameter family of Dirac Lorentz scalar potentials associated with exceptional orthogonal polynomials

We extend our recent works [ Int. J. Mod. Phys. A 38 (2023) 2350069-1] and obtain one parameter $(λ)$ family of rationally extended Dirac Lorentz scalar potentials with their explicit solutions in terms of $X_{m}$ exceptional orthogonal polynomials. We further show that as the parameter $λ\rightarrow 0$ or $-1$, we get the corresponding rationally extended Pursey and the rationally extended Abraham-Moses type of scalar potentials respectively, which have one bound state less than the starting scalar potentials.

quant-ph↗
arXiv2023

Influential Billboard Slot Selection using Spatial Clustering and Pruned Submodularity Graph

Billboard advertising is a popular out-of-home advertising technique adopted by commercial houses. Companies own billboards and offer them to commercial houses on a payment basis. Given a database of billboards with slot information, we want to determine which k slots to choose to maximize influence. We call this the INFLUENTIAL BILLBOARD SLOT SELECTION (IBSS) Problem and pose it as a combinatorial optimization problem. We show that the influence function considered in this paper is non-negative, monotone, and submodular. The incremental greedy approach based on the marginal gain computation leads to a constant factor approximation guarantee. However, this method scales very poorly when the size of the problem instance is very large. To address this, we propose a spatial partitioning and pruned submodularity graph-based approach that is divided into the following three steps: preprocessing, pruning, and selection. We analyze the proposed solution approaches to understand their time, space requirement, and performance guarantee. We conduct extensive set of experiments with real-world datasets and compare the performance of the proposed solution approaches with the available baseline methods. We observe that the proposed approaches lead to more influence than all the baseline methods within reasonable computational time.

cs.DB↗
arXiv2023

Cost and Reliability Aware Scheduling of Workflows Across Multiple Clouds with Security Constraints

Many real-world scientific workflows can be represented by a Directed Acyclic Graph (DAG), where each node represents a task and a directed edge signifies a dependency between two tasks. Due to the increasing computational resource requirements of these workflows, they are deployed on multi-cloud systems for execution. In this paper, we propose a scheduling algorithm that allocates resources to the tasks present in the workflow using an efficient list-scheduling approach based on the parameters cost, processing time, and reliability. Next, for a given a task-resource mapping, we propose a cipher assignment algorithm that assigns security services to edges responsible for transferring data in time-optimal manner subject to a given security constraint. The proposed algorithms have been analyzed to understand their time and space requirements. We implement the proposed scheduling and cipher assignment algorithm and experimented with two real-world scientific workflows namely Epigenomics and Cybershake. We compare the performance of the proposed scheduling algorithm with the state-of-art evolutionary methods. We observe that our method outperforms the state-of-art methods always in terms of cost and reliability, and is inferior in terms of makespan in some cases.

cs.DC↗
arXiv2023

Few-Shot Domain Adaptation For End-to-End Communication

The problem of end-to-end learning of a communication system using an autoencoder -- consisting of an encoder, channel, and decoder modeled using neural networks -- has recently been shown to be an effective approach. A challenge faced in the practical adoption of this learning approach is that under changing channel conditions (e.g. a wireless link), it requires frequent retraining of the autoencoder in order to maintain a low decoding error rate. Since retraining is both time consuming and requires a large number of samples, it becomes impractical when the channel distribution is changing quickly. We propose to address this problem using a fast and sample-efficient (few-shot) domain adaptation method that does not change the encoder and decoder networks. Different from conventional training-time unsupervised or semi-supervised domain adaptation, here we have a trained autoencoder from a source distribution that we want to adapt (at test time) to a target distribution using only a small labeled dataset, and no unlabeled data. We focus on a generative channel model based on the Gaussian mixture density network (MDN), and propose a regularized, parameter-efficient adaptation of the MDN using a set of affine transformations. The learned affine transformations are then used to design an optimal transformation at the decoder input to compensate for the distribution shift, and effectively present to the decoder inputs close to the source distribution. Experiments on many simulated distribution changes common to the wireless setting, and a real mmWave FPGA testbed demonstrate the effectiveness of our method at adaptation using very few target domain samples. The code for our work can be found at: https://github.com/jayaram-r/domain-adaptation-autoencoder.

cs.LG↗
arXiv2022

A Makespan and Energy-Aware Scheduling Algorithm for Workflows under Reliability Constraint on a Multiprocessor Platform

Many scientific workflows can be modeled as a Directed Acyclic Graph (henceforth mentioned as DAG) where the nodes represent individual tasks, and the directed edges represent data and control flow dependency between two tasks. Due to the large volume of data, multiprocessor systems are often used to execute these workflows. Hence, scheduling the tasks of a workflow to achieve certain goals (such as minimizing the makespan, energy, or maximizing reliability, processor utilization, etc.) remains an active area of research in embedded systems. In this paper, we propose a workflow scheduling algorithm to minimize the makespan and energy for a given reliability constraint. If the reliability constraint is higher, we further propose Energy Aware Fault Tolerant Scheduling (henceforth mentioned as EAFTS) based on active replication. Additionally, given that the allocation of task nodes to processors is known, we develop a frequency allocation algorithm that assigns frequencies to the processors. Mathematically we show that our algorithms can work for any satisfiable reliability constraint. We analyze the proposed solution approaches to understand their time requirements. Experiments with real-world Workflows show that our algorithms, MERT and EAFTS, outperform the state-of-art approaches. In particular, we observe that MERT gives 3.12% lesser energy consumption and 14.14% lesser makespan on average. In the fault-tolerant setting, our method EAFTS gives 11.11% lesser energy consumption on average when compared with the state-of-art approaches.

cs.CY↗
arXiv2022

CEDCES: A Cost Effective Deadline Constrained Evolutionary Scheduler for Task Graphs in Multi-Cloud System

Many scientific workflows can be modeled as a Directed Acyclic Graph (henceforth mentioned as DAG) where the nodes represent individual tasks and the directed edges represent data and control flow dependency between two tasks. Due to large computational resource requirements, a single cloud cannot meet the requirements of the workflow. Hence, a multi-cloud system, where multiple cloud providers pool their resources together becomes a good solution. The major objectives considered while scheduling the tasks present in a task graph include execution cost and makespan. In this paper, we present Cost Effective Deadline Constrained Evolutionary Scheduler (henceforth mentioned as CEDCES) which aims to minimize the execution cost under a given deadline constraint. CEDCES contains Particle Swarm Optimization-based (henceforth mentioned as PSO) method in its core, however includes novel initialization, crossover, and mutation schemes. Extensive simulation experiments on real-world workflows show that CEDCES outperforms the state-of-art algorithms, in particular, 60.41% on average in terms of execution cost. In cases where the deadline is violated, CEDCES gives the least overshoot in execution time and outperforming the others by 10.96% on average.

cs.DC↗
arXiv2022

A Cost Effective Reliability Aware Scheduler for Task Graphs in Multi-Cloud System

Many scientific workflows can be represented by a Directed Acyclic Graph (DAG) where each node represents a task, and there will be a directed edge between two tasks if and only if there is a dependency relationship between the two i.e. the second one can not be started unless the first one is finished. Due to the increasing computational requirements of these workflows, they are deployed on cloud computing systems. Scheduling of workflows on such systems to achieve certain goals(e.g. minimization of makespan, cost, or maximization of reliability, etc.) remains an active area of research. In this paper, we propose a scheduling algorithm for allocating the nodes of our task graph in a heterogeneous multi-cloud system. The proposed scheduler considers many practical concerns such as pricing mechanisms, discounting schemes, and reliability analysis for task execution. This is a list-based heuristic that allocates tasks based on the expected times for which VMs need to be rented for them. We have analyzed the proposed approach to understand its time requirement. We perform a large number of experiments with real-world workflows: FFT, Ligo, Epigenomics, and Random workflows and observe that the proposed scheduler outperforms the state-of-art approaches up to 12%, 11%, and 1.1% in terms of cost, makespan, and reliability, respectively.

cs.DC↗
arXiv2022

Solutions of (1+1)-dimensional Dirac equation associated with exceptional orthogonal polynomials and the parametric symmetry

We consider $1+1$-dimensional Dirac equation with rationally extended scalar potentials corresponding to the radial oscillator, the trigonometric Scarf and the hyperbolic Poschl-Teller potentials and obtain their solution in terms of exceptional orthogonal polynomials. Further, in the case of the trigonometric Scarf and the hyperbolic Poschl-Teller cases, new family of Dirac scalar potentials are generated using the idea of parametric symmetry and their solutions are obtained in terms of conventional as well as exceptional orthogonal polynomials.

math-ph↗
arXiv2022

Profit Maximization using Social Networks in Two-Phase Setting

Now-a-days, \emph{Online Social Networks} have been predominantly used by commercial houses for viral marketing where the goal is to maximize profit. In this paper, we study the problem of Profit Maximization in the two\mbox{-}phase setting. The input to the problem is a \emph{social network} where the users are associated with a cost and benefit value, and a fixed amount of budget splitted into two parts. Here, the cost and the benefit associated with a node signify its incentive demand and the amount of benefit that can be earned by influencing that user, respectively. The goal of this problem is to find out the optimal seed sets for both phases such that the aggregated profit at the end of the diffusion process is maximized. First, we develop a mathematical model based on the \emph{Independent Cascade Model} of diffusion that captures the aggregated profit in an \emph{expected} sense. Subsequently, we show that selecting an optimal seed set for the first phase even considering the optimal seed set for the second phase can be selected efficiently, is an $\textsf{NP}$-Hard Problem. Next, we propose two solution methodologies, namely the \emph{single greedy} and the \emph{double greedy} approach for our problem that works based on marginal gain computation. A detailed analysis of both methodologies has been done to understand their time and space requirements. We perform an extensive set of experiments to demonstrate the effectiveness and efficiency of the proposed approaches with real-world datasets. From the experiments, we observe that the proposed solution approaches lead to more profit compared to the baseline methods and in particular, the double greedy approach leads to up to $5 \%$ improvement compared to its single\mbox{-}phase counterpart.

cs.DB↗
arXiv2022

Utility Driven Job Selection Problem on Road Networks

In this paper, we study the problem of \textsc{Utility Driven Job Selection} on Road Networks for which the inputs are: a road network with the vertices as the set of Point-Of-Interests (Henceforth mentioned as POI) and the edges are road segments joining the POIs, a set of jobs with their originating POI, starting time, duration, and the utility. A worker can earn the utility associated with the job if (s)he performs this. As the jobs are originating at different POIs, the worker has to move from one POI to the other one to take up the job. Some budget is available for this purpose. Any two jobs can be taken up by the worker only if the finishing time of the first job plus traveling time from the POI of the first job to the second one should be less than or equal to the starting time of the second job. We call this constraint as the temporal constraint. The goal of this problem is to choose a subset of the jobs to maximize the earned utility such that the budget and temporal constraints should not be violated. We present two solution approaches with detailed analysis. First one of them works based on finding the locally optimal job at the end of every job and we call this approach as the \emph{Best First Search Approach}. The other approach is based on the Nearest Neighbor Search on road networks. We perform a set of experiments with real\mbox{-}world trajectory datasets to demonstrate the efficiency and effectiveness of the proposed solution approaches. We observe that the proposed approaches lead to more utility compared to baseline methods.

cs.DB↗
arXiv2022

Envy\mbox{-}free Trip Planning in Group Trip Planning Query Problem

In recent times, Group Trip Planning Query (henceforth referred to as GTP Query) is one of the well\mbox{-}studied problems in Spatial Databases. The inputs to the problem are a road network where the vertices represent the Point-of-Interests (mentioned as POIs henceforth) and they are grouped into different categories, edges represent the road segments, and edge weight represents the distance and a group of users along with their source and destination location. This problem asks to return one POI from every category such that the aggregated distance traveled by the group is minimized. As the objective is to minimize the aggregated distance, the existing solution methodologies do not consider the individual distances traveled by the group members. To address this issue, we introduce and study the \textsc{Envy Free Group Trip Planning Query} Problem. Along with the inputs of the GTP Query Problem, in this variant, we also have a threshold distance $D$ such that aggregated distance traveled by the group is minimized and for any member pairs the difference between their individual distance traveled is less than equal to $D$. However, it may so happen that a given $D$ value no such set POIs are found. To tackle this issue, we introduce the surrogate problem \textsc{Envy Free Group Trip Planning Query with Minimum Additional Distance} Problem which asks what is the minimum distance to be added with $D$ to obtain at least one solution. For these problems, we design efficient solution approaches and experiment with real-world datasets. From the experiments, we observe that the proposed solution approaches lead to less aggregated distance compared to baseline methods with reasonable computational overhead.

cs.DB↗
arXiv2022

Influential Billboard Slot Selection using Pruned Submodularity Graph

Billboard Advertisement has emerged as an effective out-of-home advertisement technique and adopted by many commercial houses. In this case, the billboards are owned by some companies and they are provided to the commercial houses slot\mbox{-}wise on a payment basis. Now, given the database of billboards along with their slot information which $k$ slots should be chosen to maximize the influence. Formally, we call this problem as the \textsc{Influential Billboard Slot Selection} Problem. In this paper, we pose this problem as a combinatorial optimization problem. Under the `triggering model of influence', the influence function is non-negative, monotone, and submodular. However, as the incremental greedy approach for submodular function maximization does not scale well along with the size of the problem instances, there is a need to develop efficient solution methodologies for this problem.

cs.DB↗
arXiv2022

Hierarchical Federated Learning with Privacy

Federated learning (FL), where data remains at the federated clients, and where only gradient updates are shared with a central aggregator, was assumed to be private. Recent work demonstrates that adversaries with gradient-level access can mount successful inference and reconstruction attacks. In such settings, differentially private (DP) learning is known to provide resilience. However, approaches used in the status quo (\ie central and local DP) introduce disparate utility vs. privacy trade-offs. In this work, we take the first step towards mitigating such trade-offs through {\em hierarchical FL (HFL)}. We demonstrate that by the introduction of a new intermediary level where calibrated DP noise can be added, better privacy vs. utility trade-offs can be obtained; we term this {\em hierarchical DP (HDP)}. Our experiments with 3 different datasets (commonly used as benchmarks for FL) suggest that HDP produces models as accurate as those obtained using central DP, where noise is added at a central aggregator. Such an approach also provides comparable benefit against inference adversaries as in the local DP case, where noise is added at the federated clients.

cs.LG↗