Scalable Backdoor Detection in Neural Networks

Recently, it has been shown that deep learning models are vulnerable to Trojan attacks, where an attacker can install a backdoor during training time to make the resultant model misidentify samples contaminated with a small trigger patch. Current backdoor detection methods fail to achieve good detection performance and are computationally expensive. In this paper, we propose a novel trigger reverse-engineering based approach whose computational complexity does not scale with the number of labels, and is based on a measure that is both interpretable and universal across different network and patch types. In experiments, we observe that our method achieves a perfect score in separating Trojaned models from pure models, which is an improvement over the current state-of-the art method.

Selection-based Question Answering of an MOOC

e-Yantra Robotics Competition (eYRC) is a unique Robotics Competition hosted by IIT Bombay that is actually an Embedded Systems and Robotics MOOC. Registrations have been growing exponentially in each year from 4500 in 2012 to over 34000 in 2019. In this 5-month long competition students learn complex skills under severe time pressure and have access to a discussion forum to post doubts about the learning material. Responding to questions in real-time is a challenge for project staff. Here, we illustrate the advantage of Deep Learning for real-time question answering in the eYRC discussion forum. We illustrate the advantage of Transformer based contextual embedding mechanisms such as Bidirectional Encoder Representation From Transformer (BERT) over word embedding mechanisms such as Word2Vec. We propose a weighted similarity metric as a measure of matching and find it more reliable than Content-Content or Title-Title similarities alone. The automation of replying to questions has brought the turn around response time(TART) down from a minimum of 21 mins to a minimum of 0.3 secs.

Enhancing Speech Intelligibility in Text-To-Speech Synthesis using Speaking Style Conversion

The increased adoption of digital assistants makes text-to-speech (TTS) synthesis systems an indispensable feature of modern mobile devices. It is hence desirable to build a system capable of generating highly intelligible speech in the presence of noise. Past studies have investigated style conversion in TTS synthesis, yet degraded synthesized quality often leads to worse intelligibility. To overcome such limitations, we proposed a novel transfer learning approach using Tacotron and WaveRNN based TTS synthesis. The proposed speech system exploits two modification strategies: (a) Lombard speaking style data and (b) Spectral Shaping and Dynamic Range Compression (SSDRC) which has been shown to provide high intelligibility gains by redistributing the signal energy on the time-frequency domain. We refer to this extension as Lombard-SSDRC TTS system. Intelligibility enhancement as quantified by the Intelligibility in Bits (SIIB-Gauss) measure shows that the proposed Lombard-SSDRC TTS system shows significant relative improvement between 110% and 130% in speech-shaped noise (SSN), and 47% to 140% in competing-speaker noise (CSN) against the state-of-the-art TTS approach. Additional subjective evaluation shows that Lombard-SSDRC TTS successfully increases the speech intelligibility with relative improvement of 455% for SSN and 104% for CSN in median keyword correction rate compared to the baseline TTS method.

L^2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks

Graph convolution networks (GCN) are increasingly popular in many applications, yet remain notoriously hard to train over large graph datasets. They need to compute node representations recursively from their neighbors. Current GCN training algorithms suffer from either high computational costs that grow exponentially with the number of layers, or high memory usage for loading the entire graph and node embeddings. In this paper, we propose a novel efficient layer-wise training framework for GCN (L-GCN), that disentangles feature aggregation and feature transformation during training, hence greatly reducing time and memory complexities. We present theoretical analysis for L-GCN under the graph isomorphism framework, that L-GCN leads to as powerful GCNs as the more costly conventional training algorithm does, under mild conditions. We further propose L^2-GCN, which learns a controller for each layer that can automatically adjust the training epochs per layer in L-GCN. Experiments show that L-GCN is faster than state-of-the-arts by at least an order of magnitude, with a consistent of memory usage not dependent on dataset size, while maintaining comparable prediction performance. With the learned controller, L^2-GCN can further cut the training time in half. Our codes are available at https://github.com/Shen-Lab/L2-GCN and supplementary materials at https://slack-files.com/TC7R2EBMJ-F012C60T335-281aabd097.

Cascaded Text Generation with Markov Transformers

The two dominant approaches to neural text generation are fully autoregressive models, using serial beam search decoding, and non-autoregressive models, using parallel decoding with no output dependencies. This work proposes an autoregressive model with sub-linear parallel time generation. Noting that conditional random fields with bounded context can be decoded in parallel, we propose an efficient cascaded decoding approach for generating high-quality output. To parameterize this cascade, we introduce a Markov transformer, a variant of the popular fully autoregressive model that allows us to simultaneously decode with specific autoregressive context cutoffs. This approach requires only a small modification from standard autoregressive training, while showing competitive accuracy/speed tradeoff compared to existing methods on five machine translation datasets.

One-Bit Compressed Sensing via One-Shot Hard Thresholding

This paper concerns the problem of 1-bit compressed sensing, where the goal is to estimate a sparse signal from a few of its binary measurements. We study a non-convex sparsity-constrained program and present a novel and concise analysis that moves away from the widely used notion of Gaussian width. We show that with high probability a simple algorithm is guaranteed to produce an accurate approximation to the normalized signal of interest under the $\ell_2$-metric. On top of that, we establish an ensemble of new results that address norm estimation, support recovery, and model misspecification. On the computational side, it is shown that the non-convex program can be solved via one-step hard thresholding which is dramatically efficient in terms of time complexity and memory footprint. On the statistical side, it is shown that our estimator enjoys a near-optimal error rate under standard conditions. The theoretical results are substantiated by numerical experiments.

Conv-Transformer Transducer: Low Latency, Low Frame Rate, Streamable End-to-End Speech Recognition

Transformer has achieved competitive performance against state-of-the-art end-to-end models in automatic speech recognition (ASR), and requires significantly less training time than RNN-based models. The original Transformer, with encoder-decoder architecture, is only suitable for offline ASR. It relies on an attention mechanism to learn alignments, and encodes input audio bidirectionally. The high computation cost of Transformer decoding also limits its use in production streaming systems. To make Transformer suitable for streaming ASR, we explore Transducer framework as a streamable way to learn alignments. For audio encoding, we apply unidirectional Transformer with interleaved convolution layers. The interleaved convolution layers are used for modeling future context which is important to performance. To reduce computation cost, we gradually downsample acoustic input, also with the interleaved convolution layers. Moreover, we limit the length of history context in self-attention to maintain constant computation cost for each decoding step. We show that this architecture, named Conv-Transformer Transducer, achieves competitive performance on LibriSpeech dataset (3.6\% WER on test-clean) without external language models. The performance is comparable to previously published streamable Transformer Transducer and strong hybrid streaming ASR systems, and is achieved with smaller look-ahead window (140~ms), fewer parameters and lower frame rate.

Mining self-similarity: Label super-resolution with epitomic representations

We show that simple patch-based models, such as epitomes, can have superior performance to the current state of the art in semantic segmentation and label super-resolution, which uses deep convolutional neural networks. We derive a new training algorithm for epitomes which allows, for the first time, learning from very large data sets and derive a label super-resolution algorithm as a statistical inference algorithm over epitomic representations. We illustrate our methods on land cover mapping and medical image analysis tasks.

IDOL: A Framework for IMU-DVS Odometry using Lines

In this paper, we introduce IDOL, an optimization-based framework for IMU-DVS Odometry using Lines. Event cameras, also called Dynamic Vision Sensors (DVSs), generate highly asynchronous streams of events triggered upon illumination changes for each individual pixel. This novel paradigm presents advantages in low illumination conditions and high-speed motions. Nonetheless, this unconventional sensing modality brings new challenges to perform scene reconstruction or motion estimation. The proposed method offers to leverage a continuous-time representation of the inertial readings to associate each event with timely accurate inertial data. The method's front-end extracts event clusters that belong to line segments in the environment whereas the back-end estimates the system's trajectory alongside the lines' 3D position by minimizing point-to-line distances between individual events and the lines' projection in the image space. A novel attraction/repulsion mechanism is presented to accurately estimate the lines' extremities, avoiding their explicit detection in the event data. The proposed method is benchmarked against a state-of-the-art frame-based visual-inertial odometry framework using public datasets. The results show that IDOL performs at the same order of magnitude on most datasets and even shows better orientation estimates. These findings can have a great impact on new algorithms for DVS.

A Sensitivity Analysis Approach for Evaluating a Radar Simulation for Virtual Testing of Autonomous Driving Functions

Simulation-based testing is a promising approach to significantly reduce the validation effort of automated driving functions. Realistic models of environment perception sensors such as camera, radar and lidar play a key role in this testing strategy. A generally accepted method to validate these sensor models does not yet exist. Particularly radar has traditionally been one of the most difficult sensors to model. Although promising as an alternative to real test drives, virtual tests are time-consuming due to the fact that they simulate the entire radar system in detail, using computation-intensive simulation techniques to approximate the propagation of electromagnetic waves. In this paper, we introduce a sensitivity analysis approach for developing and evaluating a radar simulation, with the objective to identify the parameters with the greatest impact regarding the system under test. A modular radar system simulation is presented and parameterized to conduct a sensitivity analysis in order to evaluate a spatial clustering algorithm as the system under test, while comparing the output from the radar model to real driving measurements to ensure a realistic model behavior. The presented approach is evaluated and it is demonstrated that with this approach results from different situations can be traced back to the contribution of the individual sub-modules of the radar simulation.