We present Structured World Modeling for Policy Optimization (SWMPO), a framework for unsupervised learning of neurosymbolic Finite State Machines (FSM) that capture environmental structure for policy optimization. Traditional unsupervised world modeling methods rely on unstructured representations, such as neural networks, that do not explicitly represent high-level patterns within the system (e.g., patterns in the dynamics of regions such as \emph{water} and \emph{land}). Instead, SWMPO models the environment as a finite state machine (FSM), where each state corresponds to a specific region with distinct dynamics. This structured representation can then be leveraged for tasks like policy optimization. Previous works that synthesize FSMs for this purpose have been limited to discrete spaces, not continuous spaces. Instead, our proposed FSM synthesis algorithm operates in an unsupervised manner, leveraging low-level features from unprocessed, non-visual data, making it adaptable across various domains. The synthesized FSM models are expressive enough to be used in a model-based Reinforcement Learning scheme that leverages offline data to efficiently synthesize environment-specific world models. We demonstrate the advantages of SWMPO by benchmarking its environment modeling capabilities in simulated environments.

Providing resources and services from various cloud providers is now an increasingly promising paradigm. Workflow applications are becoming increasingly computation-intensive or data-intensive, with resource allocation being maintained in terms of pay per usage. In this paper, a multi-objective optimization study for scientific workflow in a cloud environment is proposed. The aim is to minimize execution time and purchasing cost simultaneously while satisfying the demand requirements of customers. The uncertainties present in the model are identified and handled using a well-known technique called Chance Constrained Programming (CCP) for real-world implementation. The model is solved using the Non-dominated Sorting Genetic Algorithm – II (NSGA-II). This comprehensive study shows that the solutions obtained on considering uncertainties vary from the deterministic case. Based on the probability of constraint satisfaction, the objective functions improve but at the cost of reliability of the solution. Copyright © 2021 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

Existing approaches for open-domain question answering (QA) are typically designed for questions that require either single-hop or multi-hop reasoning, which make strong assumptions of the complexity of questions to be answered. Also, multi-step document retrieval often incurs higher number of relevant but non-supporting documents, which dampens the downstream noise-sensitive reader module for answer extraction. To address these challenges, we propose a unified QA framework to answer any-hop open-domain questions, which iteratively retrieves, reranks and filters documents, and adaptively determines when to stop the retrieval process. To improve the retrieval accuracy, we propose a graph-based reranking model that perform multi-document interaction as the core of our iterative reranking framework. Our method consistently achieves performance comparable to or better than the state-of-the-art on both single-hop and multi-hop open-domain QA datasets, including Natural Questions Open, SQuAD Open, and HotpotQA.

Recent studies on open-domain question answering have achieved prominent performance improvement using pre-trained language models such as BERT. State-of-the-art approaches typically follow the "retrieve and read" pipeline and employ BERT-based reranker to filter retrieved documents before feeding them into the reader module. The BERT retriever takes as input the concatenation of question and each retrieved document. Despite the success of these approaches in terms of QA accuracy, due to the concatenation, they can barely handle high-throughput of incoming questions each with a large collection of retrieved documents. To address the efficiency problem, we propose DC-BERT, a decoupled contextual encoding framework that has dual BERT models: an online BERT which encodes the question only once, and an offline BERT which pre-encodes all the documents and caches their encodings. On SQuAD Open and Natural Questions Open datasets, DC-BERT achieves 10x speedup on document retrieval, while retaining most (about 98%) of the QA performance compared to state-of-the-art approaches for open-domain question answering.

Due to spatio-temporal variation of mobile subscriber's data traffic requirements, traffic load experienced by base stations present at different cell sites exhibit highly dynamic behavior in traditional cellular systems. This non-uniform and dynamic traffic load leads to under utilization of the base station computing resources at cell sites. Cloud Radio Access Network (C-RAN) is an innovative architecture which addresses this issue and keeps the Total Cost of Ownership (TCO) under safe limit for cellular operators. In C-RAN, the baseband processing units (BBUs) are segregated from cell sites and are pooled in a central cloud data center thereby facilitating shared access for a set of Remote Radio Heads (RRHs) present at cell sites. In order to truly exploit the benefits of C-RAN, the BBU pool deployed in the cloud has to efficiently serve clusters of RRHs (i.e., many-to-one mapping between RRHs and BBUs in the BBU pool) and thereby minimizing the required number of active BBUs. In this work, potential benefits of C-RAN are studied by considering realistic traffic loads of base stations deployed in urban areas by using statistical models. We propose a lightweight and load-aware algorithm, Dynamic RRH Assignment (DRA), which achieves BBU pooling gain close to that of a well known First-Fit Decreasing (FFD) bin packing algorithm. Using extensive simulations, we show that DRA consumes only 25% of time on average compared to FFD for the case of urban cellular deployment of 1000 RRHs. DRA slightly overestimates the required number of active BBUs as compared to FFD by 1.7% and 1.4% for weekdays and weekends, respectively.

Femto cells a.k.a. Low Power Nodes (LPNs) are used to improve indoor data rates as well as to reduce traffic load on macro Base Stations (BSs) in LTE cellular networks. These LPNs are deployed inside office buildings and residential apartment complexes to provide high data rates to indoor Users. With high SINR (Signal-to-Interference plus Noise Ratio) the users experience good throughput, but the SINR decreases significantly because of interference and obstacles such as building walls, present in the communication path. So, efficient placement of Femtos in buildings while considering Macro-Femto interference is very crucial for attaining desirable SINR. At the same time, minimizing the power leakage in order to improve the signal strength of outdoor users in a high interference (HIZone) around the building area is important. In our work, we have considered obstacles (walls, floors) and interference between Macro and Femto BSs. To be fair to both indoor and outdoor users, we designed an efficient placement and power control SON (Self organizing Network) algorithm which optimally places Femtos and dynamically adjusts the transmission power of Femtos based on the occupancy of Macro users in the HIZone. To do this, we solve two Mixed Integer Programming (MIP) methods namely: Minimize number of Femtos (MinNF) method which guarantees threshold SINR (SINRTh) -2dB for all indoor users and optimal Femto power (OptFP) allocation method which guarantees SINRTh(- 4 dB) for indoor users with the Macro users SINR degradation as lesser than 2dB.