Protein-Protein Interactions (PPIs) are a crucial mechanism underpinning the function of the cell. So far, a wide range of machine-learning based methods have been proposed for predicting these relationships. Their success is heavily dependent on the construction of the underlying feature vectors, with most using a set of physico-chemical properties derived from the sequence. Few work directly with the sequence itself. In this paper, we explore the utility of sequence embeddings for predicting protein-protein interactions. We construct a protein pair feature vector by concatenating the embeddings of their constituent sequence. These feature vectors are then used as input to a binary classifier to make predictions. To learn sequence embeddings, we use two established Word2Vec based methods - Seq2Vec and BioVec - and we also introduce a novel feature construction method called SuperVecNW. The embeddings generated through SuperVecNW capture some network information in addition to the contextual information present in the sequences. We test the efficacy of our proposed approach on human and yeast PPI datasets and on three well-known networks: CD9, the Ras-Raf-Mek-Erk-Elk-Srf pathway, and a Wnt-related network. We demonstrate that low dimensional sequence embeddings provide better results than most alternative representations based on physico-chemical properties while offering a far simple approach to feature vector construction.

With the increased proliferation of smart devices, the transit passengers of today expect a higher quality of service in the form of real-time traffic updates, accurate expected time-of-arrival (ETA) predictions. Providing these services requires public transit agencies and private transportation players to maintain full situational awareness of the city-wide traffic. However, most such agencies and companies are resource-constrained and do not have access to city-wide traffic data. The availability of sparsely sampled and outlier-corrupted traffic data renders the resulting traffic maps patchy and unreliable and necessitates the use of sophisticated real-time traffic interpolation and prediction algorithms. Moreover, since the traffic data is measured and collected in a sequential manner, the estimations must also be generated online. Thankfully, the traffic matrices are spatially and temporally structured, allowing the use of time-series and matrix/tensor completion algorithms. This work puts forth a generative model for the traffic density and subsequently uses a variational Bayesian formalism to learn the parameters of the model. Specifically, we consider low-rank traffic matrices whose subspace evolves according to a state-space model with possible sparse outliers. Unlike most matrix/tensor completion algorithms, the proposed model is equipped with automatic relevance determination priors that allow it to learn the parameters in an entirely data-driven manner. A forward-backward algorithm is proposed that enables the updates to be carried out at low-complexity. Simulations carried out on real traffic speed data demonstrate that the proposed algorithm better predicts the future traffic densities as compared to the state-of-the-art matrix/tensor completion algorithms.

Many real world applications that suffer from missing data and outliers can be modeled in a matrix completion framework. In this paper, we consider low-rank matrices whose subspace evolves according to a state-space model and propose an online variational Bayesian formulation to learn the low rank components as well as the state-space model. Unlike the other matrix/tensor completion techniques, in our framework, the key algorithm parameters like rank and various noise power need not be fine-tuned and are learned automatically. We also propose a forward-backward algorithm that allows update to be carried out at low complexity manner. Simulations performed on the real world traffic data illustrates promising imputation as well as temporal prediction performance even in an online setup.

One of the key challenges in G.fast is to minimize the power consumption at distribution points. G.fast standards define Discontinuous Operation modes that provide avenues for power reduction by allowing intermittent transmission of users along time slots. In this paper, we formulate a power efficiency problem as a user-slot assignment problem, where we schedule users to time slots during discontinuous operation such that the total power consumption is minimized. Since the general user-slot assignment is NP hard, we propose a weighted$A^{*}$algorithm based solution that achieves reasonable performance with limited computational resources. Further, the proposed user-slot assignment is also G.fast standards compliant and therefore can be implemented in practice. We also explore different precoding as well as user grouping strategies that can be employed while performing user-slot allocation. Finally, the main insight of this work is that on using our user-slot assignment algorithm and by only precoding during the normal operations, we achieve energy efficiency levels comparable to those achieved when precoding is applied during the discontinuous operations without the suggested user-slot allocation.

Many applications of single channel source separation (SCSS) including automatic speech recognition (ASR), hearing aids etc. require an estimation of only one source from a mixture of many sources.Treating this special case as a regular SCSS problem where in all constituent sources are given equal priority in terms of reconstruction may result in a suboptimal separation performance.In this paper, we tackle the one source separation problem by suitably modifying the orthodox SCSS framework and focus only on one source at a time.The proposed approach is a generic framework that can be applied to any existing SCSS algorithm, improves performance, and scales well when there are more than two sources in the mixture unlike most existing SCSS methods.Additionally, existing SCSS algorithms rely on fine hyper-parameter tuning hence making them difficult to use in practice.Our framework takes a step towards automatic tuning of the hyper-parameters thereby making our method better suited for the mixture to be separated and thus practically more useful.We test our framework on a neural network based algorithm and the results show an improved performance in terms of SDR and SAR.

The adoption of precoding (vectoring) in VDSL2 at the central office has resulted in mitigation of the far-end crosstalk seen at the customer premises equipment (CPE). As a result, alien noise (including repetitive impulse noise) is the new dominant source of impairment for downstream VDSL. At the CPE, an additional common mode (CM) sensor can sense the electromagnetically coupled alien noise signal, which can then be used to cancel the alien noise coupling into the differential mode signal. The intermittent and repetitive nature of the noise sources necessitates that the CM sensor based noise cancellation algorithm be capable of training and adapting during data mode in the presence of the useful data signal, since the presence of alien noise cannot be guaranteed during the modem training phase. In this paper, we propose a novel two-stage frequency domain algorithm based on a per-tone cancellation model for this purpose. The proposed algorithm outperforms previously proposed time domain algorithms in terms of convergence speed in many practical scenarios due to the decision directed nature of the proposed algorithm. We also analyse the theoretical convergence of the algorithm, which is also validated by the simulation experiments.

For VDSL systems, alien noise cancellation using an additional common mode sensor at the customer premises equipment (CPE) receiver can be done by combining the differential-mode (DM) signal with the common-mode (CM) signal passed through a long linear filter. Frequency domain per-tone cancellation offers a low complexity approach to the problem but suffers from loss in cancellation performance due to approximations in the per-tone model. We analyze this loss and show that it is possible to minimize it by a post-training “delay” adjustment. We also address the problem of training such a noise canceller during data mode in the presence of a much stronger useful data signal in DM since noise events may not occur during modem train-up. We propose an algorithm based on the per-tone approach which is capable of fast convergence during data mode for intermittent alien noise sources, analyze its convergence behaviour and demonstrate the usefulness of the pertone approach and the proposed training algorithm over existing time domain methods.

To enable energy efficiency, G.fast standards define discontinuous operations (DO) where a set of users can remain inactive while others transmit during a time domain duplex (TDD) frame. In this work, we investigate energy efficient discontinuous operations (DO) by scheduling users to time slots, such that the total energy consumption is minimized while satisfying the individual data rate constraints. Since the user-slot assignment problem is NP complete in nature, we propose the use of weighted A* algorithm that achieves reasonable performance with limited computational resources. The main insight of this work is that on using our user-slot assignment algorithm and by only precoding during the normal operations, we achieve the same energy efficiency as achieved by precoding strategies like discontinuous vectoring [1] while satisfying the provisions of the G.fast standard.

While the density of access points in enterprise settings has increased, the sharing of the spatial resource amongst links in 802.11 wireless local area networks remains inefficient. Conservative mechanisms based on a static carrier sense range (CSR) are used and are designed to avoid occurrence of interfering transmissions. Even when the CSR is adapted to allow interfering transmissions, it is with the goal of increasing spatial reuse, which may not translate to a larger network throughput. We formulate the network throughput optimization problem, which is to decide which links in a network must share in space (transmit data simultaneously) such that the network throughput is maximized. Links share in space by piggybacking on data transmission opportunities seized by another link using RTS/CTS as specified in the distributed coordination function (DCF) of 802.11. Sharing in space increases interference and hence reduces the PHY rate at which a link can send data. It also increases the opportunities a link gets to transmit data, however. The optimization problem is NP hard. A relaxation of the problem gives an upper bound on network throughput. We propose computationally feasible algorithms that achieve a significant percentage of the upper bound. Our network modeling and evaluation is restricted to 802.11 networks in which all nodes always have a packet to send and are within carrier sense range of each other. Networks with a high density of clients and AP(s) are shown, via simulation, to achieve large throughput gains (up to 400% for 25 clients and AP(s)), over standard 802.11.

Alien noise in the vectored very-high-speed digital subscriber line (VDSL) system is part of the additive noise at the receiver and exhibits strong correlation among users. We present a per-tone co-operative alien noise cancellation (CoMAC) algorithm for the upstream (US) VDSL that can be applied subsequent to any self far-end-crosstalk (FEXT) mitigation strategy. CoMAC operates by predicting the noise seen by a given user based on the error samples from the remaining users. These errors are conveniently obtained after slicing the self-FEXT canceled signal of all the vectored users. We show that if the estimation of these errors is accurate, the proposed alien canceler achieves the Cramer-Rao lower bound (CRLB). In practice, the seamless rate adaptation (SRA) operation, which enables increased bit rate by increasing the bit-loading per-tone, can cause decision errors in any decision directed strategy. We also analyze the impact of these decision errors - an issue not addressed in the literature. We propose a strategy for bit-loading during the SRA operation by formulating a max-min optimization problem and demonstrate a possibility of a guaranteed (minimum) improvement in the per-user rate. Simulations indicate that performance of the algorithm can exceed the minimum value significantly in practical situations.

We present cooperative MIMO for alien noise cancellation (CoMAC): a per-tone, blind, low-complexity, linear, and adaptive noise whitening algorithm for alien crosstalk mitigation in upstream vectored VDSL systems. CoMAC directly acts on the residual errors of the vectored users after self-FEXT cancellation and frequency domain equalization, and thus, leverages the inherent alien-crosstalk-induced spatial correlation across users. CoMAC employs a low-complexity recursion scheme derived from the optimal MMSE noise whitener to non-disruptively initialize, engage, and adapt the noise canceller while the vectored users operate in data mode. Assuming reliable transmit symbol estimation at its input, we show that CoMAC achieves the Cramer-Rao lower bound. Further, the SNR improvements accruing from CoMAC can be translated into substantial rate improvements for upstream vectored VDSL.

This paper introduces a new family of per frequency linear pre-coders, called off-diagonal MIMO pre-coders (ODMP), for discrete multi-tone (DMT) based digital subscriber line (DSL) systems. ODMP target pre-compensation of the off- diagonal self far-end cross talk (FEXT) terms in the downstream DSL-MIMO channel matrix. We derive a low complexity and adaptive ODMP that simultaneously maximizes the Shannon Capacities per tone, experienced by the (non-cooperative) end-users. The updating scheme makes use of the error samples computed in each receiver. No matrix inversion is required in the algorithm which may be used for learning, tracking, and incrementing the ODMP without any disruption of data service. Finite precision simulation results reveal that coding the complex error with 16 bits allows for the algorithm to reach "self-FEXT free" performance after only 150 iterations in the learning phase and 450 iterations in the incrementing phase.

We reexamine the problems of computer-aided classification and pairing of human chromosomes, and propose to jointly optimize the solutions of these two related problems. The combined problem is formulated into one of optimal three-dimensional assignment with an objective function of maximum likelihood. This formulation poses two technical challenges: 1) estimation of the posterior probability that two chromosomes form a pair and the pair belongs to a class and 2) good heuristic algorithms to solve the three-dimensional assignment problem which is NP-hard. We present various techniques to solve these problems. We also generalize our algorithms to cases where the cell data are incomplete as often encountered in practice.