Abstract

Vision-and-language navigation (VLN) aims to enable embodied agents to navigate in realistic environments us- ing natural language instructions. Given the scarcity of domain-specific training data and the high diversity of im- age and language inputs, the generalization of VLN agents to unseen environments remains challenging. Recent meth- ods explore pretraining to improve generalization, however, the use of generic image-caption datasets or existing small- scale VLN environments is suboptimal and results in limited improvements. In this work, we introduce BnB1, a large- scale and diverse in-domain VLN dataset. We first col- lect image-caption (IC) pairs from hundreds of thousands of listings from online rental marketplaces. Using IC pairs we next propose automatic strategies to generate millions of VLN path-instruction (PI) pairs. We further propose a shuffling loss that improves the learning of temporal or- der inside PI pairs. We use BnB to pretrain our Airbert2 model that can be adapted to discriminative and genera- tive settings and show that it outperforms state of the art for Room-to-Room (R2R) navigation and Remote Referring Ex- pression (REVERIE) benchmarks. Moreover, our in-domain pretraining significantly increases performance on a chal- lenging few-shot VLN evaluation, where we train the model only on VLN instructions from a few houses

Abstract

The visual world naturally exhibits an imbalance in the number of object or scene instances resulting in a long-tailed distribution. This imbalance poses significant challenges for classification models based on deep learning. Oversampling instances of the tail classes attempts to solve this imbalance. However, the limited visual diver- sity results in a network with poor representation ability. A simple counter to this is decoupling the representation and classifier net- works and using oversampling only to train the classifier. In this paper, instead of repeatedly re-sampling the same im- age (and thereby features), we explore a direction that attempts to generate meaningful features by estimating the tail category’s distribution. Inspired by ideas from recent work on few-shot learn- ing [ 53], we create calibrated distributions to sample additional features that are subsequently used to train the classifier. Through several experiments on the CIFAR-100-LT (long-tail) dataset with varying imbalance factors and on mini-ImageNet-LT (long-tail), we show the efficacy of our approach and establish a new state-of- the-art. We also present a qualitative analysis of generated features using t-SNE visualizations and analyze the nearest neighbors used to calibrate the tail class distributions. Our code is available at https://github.com/rahulvigneswaran/TailCalibX.

Abstract

Facilitating the discovery of drugs by combining diverse compounds is becoming prevalent, especially for treating complex diseases like cancers and HIV. A drug is a chemi- cal compound structure and any sub-structure of a chemical compound is designated as a fragment. A chemical compound or a fragment can be modeled as a graph structure. Given a set of chemical compounds and their corresponding large set of fragments modeled as graph structures, we address the problem of identifying potential combinations of diverse chemical compounds, which cover a certain percentage of the set of fragments. In this regard, the key contributions of this work are three-fold: First, we introduce the notion of Graph Transactional Coverage Patterns (GTCPs) for any given graph transactional dataset. Second, we propose an efficient model and framework for extracting GTCPs from a given graph transactional dataset. Third, we conduct an extensive performance study using three real datasets to demonstrate that it is indeed feasible to efficiently extract GTCPs using our proposed GTCP-extraction framework. We also demonstrate the effectiveness of the GTCP-extraction framework through a case study in computer-aided drug design. Index Terms—Graph mining, Graph transactions, Coverage patterns, Drug discovery

Abstract

Recently, Forbes, Kumar and Saptharishi [CCC, 2016] proved that there exists an explicit d O(1) - variate and degree d polynomial Pd ∈ VNP such that if any depth four circuit C of bounded formal degree d which computes a polynomial of bounded individual degree O(1), that is functionally equivalent to Pd, then C must have size 2 Ω(√ d log d) . The motivation for their work comes from Boolean Circuit Complexity. Based on a characterization for ACC0 circuits by Yao [FOCS, 1985] and Beigel and Tarui [CC, 1994], Forbes, Kumar and Saptharishi [CCC, 2016] observed that functions in ACC0 can also be computed by algebraic Σ∧ΣΠ circuits (i.e., circuits of the form – sums of powers of polynomials) of 2 logO(1) n size. Thus they argued that a 2 ω(poly log n) “functional” lower bound for an explicit polynomial Q against Σ∧ΣΠ circuits would imply a lower bound for the “corresponding Boolean function” of Q against non-uniform ACC0 . In their work, they ask if their lower bound be extended to Σ∧ΣΠ circuits. In this paper, for large integers n and d such that ω(log2 n) ≤ d ≤ n 0.01, we show that any Σ∧ΣΠ circuit of bounded individual degree at most O d k2
that functionally computes Iterated Matrix Multiplication polynomial IMMn,d (∈ VP) over {0, 1} n 2d must have size n Ω(k) . Since Iterated Matrix Multiplication IMMn,d over {0, 1} n 2d is functionally in GapL, improvement of the afore mentioned lower bound to hold for quasipolynomially large values of individual degree would imply a fine-grained separation of ACC0 from GapL. For the sake of completeness, we also show a syntactic size lower bound against any Σ∧ΣΠ circuit computing IMMn,d (for the same regime of d) which is tight over large fields. Like Forbes, Kumar and Saptharishi [CCC, 2016], we too prove lower bounds against circuits of bounded formal degree which functionally compute IMMn,d, for a slightly larger range of individual degree. 2012 ACM Subject Classification Theory of computation → Circuit complexity; Theory of computation → Algebraic complexity theory Keywords and phrases Functional Lower

Abstract

The Internet of Things (IoT) plays a key role in real-time monitoring at different stages of the power generation system, assisting to achieve better efficiency, minimize load on the grids by analysing usage patterns, provide faster resolutions to power outages, and so on. In this paper, we present a novel energy monitoring approach employing LoRaWAN-enabled smart energy meters and a oneM2M-based platform for collecting and analysing the data. The energy meters transmit data at 15-minutes intervals, i.e., 96 data points per day. A novel format has been developed for the LoRaWAN Protocol Data Unit (PDU) to transmit the values of phase currents and voltages, and data related to power and energy comsumption. This results in a high-resolution dataset containing more than 10,000 instances per meter, accumulated over the last four months. The data can be visualised in a live dashboard enabling the signal parameters such as Received Signal Strength Index (RSSI) and Signal to Noise Ratio (SNR) to be monitored in addition to the electrical parameters, to ensure proper data transmission. Finally, the trends in power and energy consumption of the load have been analysed, which can result in improved efficiency of building management, and early detection of electrical faults and failures

Abstract

Most of the sensors that measure physical variables such as humidity, strain, motion, temperature, and pressure are designed based on the variation of capacitance or the resistance of a circuit entity. This paper presents a 12-bit combined resistance and capacitance to digital converter (RCDC) to analyze and process the information from these sensors. The read-out circuit first converts the capacitance or resistance to an analog voltage, followed by conversion to digital data using a successive approximation (SAR) type analog to digital converter (ADC). The circuit has been designed based on the 180 nm CMOS technology. The capacitance to digital converter (CDC) circuit covers a wide range of capacitance from 30 pF to 100 pF with a 17 fF resolution, whereas the resistance to digital converter (RDC) circuit covers a range of resistance from 1 Ω to 30 Ω and has a resolution of 7 mΩ. The 12-bit CDC and RDC have a measurement time of 20 μs and show a non-linearity of 0.66% and 0.57%, respectively. Index Terms—analog to digital converter; capacitance to dig- ital converter; resistance to digital converter; read-out circuit; successive approximation

Abstract

Detection of color change is an important tool used in several analytical chemistry processes in various industries such as biomedical, diagnostics, agriculture, food processing, and so on. In this study, we present an automated color detection system to perform qualitative analysis of liquid reagents. Many of the available techniques for colorimetric applications make use of expensive, high-resolution digital cameras. We have developed a simple low-cost system based on a light emitting diode (LED) and light-dependent resistor (LDR) pair to obtain information about the color of a liquid reagent. This system has been compared with a camera-based system for color detection. The LED-LDR system provides information about the color of an object by obtaining relative intensities of red, green, and blue (RGB) light reflected by the object in a sequential process. The system was tested by obtaining color information for different concentrations of potassium permanganate solution and by determining the endpoint of a titration experiment. This versatile system can be used for many other colorimetric measurements, for example, the amount of dissolved ammonia in water using Nesslers reagent (potassium tetraiodomercurate).

Abstract

We address a distributed adaptive synchronization problem for complex networks composed of nonlinear nodes under state-dependent a priori interconnections, i.e. interconnection terms acting before control design. The interconnection terms are uncertain and the heterogeneous dynamics of the network nodes further contain state-dependent uncertainty and uncertain input matrix gain. Adaptive distributed control laws are proposed to tackle such an unsolved design. The proposed controller is verified in simulation via a multi-area load frequency control for power systems.

Abstract

The focus of this paper is (i) to derive a suitable dynamic model for a self-reconfigurable pavement sweeping robot PANTHERA and (ii) to design a robust controller for the same to tackle uncertainties stemming from the reconfiguration process, external disturbances and from actuator saturation. To meet the first objective, an Euler-Lagrangian dynamic model is proposed to incorporate the effects of configuration changes on the system dynamics. Based on this model, the second objective is met via designing a singular perturbation based robust controller which can tackle the aforementioned uncertainties without violating the actuation limits. To circumvent the vulnerability toward actuator saturation, the proposed controller is built on contraction theory, which, compared to a conventional Lyapunov theory based design, allows to improve closed-loop tracking performance without reducing the singular perturbation parameter. Experimental results on the PANTHERA reconfigurable robot validate the effectiveness of the proposed controller over the state of the art.

Abstract

Slide presentations are an effective and efficient tool used by the teaching community for classroom communication. However, this teaching model can be challenging for the blind and visually impaired (vi) students. The vi student required a personal human assistance for understand the presented slide. This shortcoming motivates us to de- sign a Classroom Slide Narration System (csns) that generates audio descriptions corresponding to the slide content. This problem poses as an image-to-markup language generation task. The initial step is to ex- tract logical regions such as title, text, equation, figure, and table from the slide image. In the classroom slide images, the logical regions are distributed based on the location of the image. To utilize the location of the logical regions for slide image segmentation, we propose the ar- chitecture, Classroom Slide Segmentation Network (cssn). The unique attributes of this architecture differs from most other semantic segmenta- tion networks. Publicly available benchmark datasets such as WiSe and SPaSe are used to validate the performance of our segmentation architec- ture. We obtained 9.54% segmentation accuracy improvement in WiSe dataset. We extract content (information) from the slide using four well- established modules such as optical character recognition (ocr), figure classification, equation description, and table structure recognizer. With this information, we build a Classroom Slide Narration System (csns) to help vi students understand the slide content. The users have given better feedback on the quality output of the proposed csns in compar- ison to existing systems like Facebook’s Automatic Alt-Text (aat) and Tesseract.