Abstract
Monocular 3D object detection is an essential perception task for autonomous driving. However, the high reliance on large-scale labeled data make it costly and time-consuming during model optimization. To reduce such over-reliance on human annotations, we propose Mix-Teaching, an effective semi-supervised learning framework applicable to employ both labeled and unlabeled images in training stage. Mix-Teaching first generates pseudo-labels for unlabeled images by self-training. The student model is then trained on the mixed images possessing much more intensive and precise labeling by merging instance-level image patches into empty backgrounds or labeled images. This is the first to break the image-level limitation and put high-quality pseudo labels from multi frames into one image for semi-supervised training. Besides, as a result of the misalignment between confidence score and localization quality, it's hard to discriminate high-quality pseudo-labels from noisy predictions using only confidence-based criterion. To that end, we further introduce an uncertainty-based filter to help select reliable pseudo boxes for the above mixing operation. To the best of our knowledge, this is the first unified SSL framework for monocular 3D object detection. Mix-Teaching consistently improves MonoFlex and GUPNet by significant margins under various labeling ratios on KITTI dataset. For example, our method achieves around +6.34% AP@0.7 improvement against the GUPNet baseline on validation set when using only 10% labeled data. Besides, by leveraging full training set and the additional 48K raw images of KITTI, it can further improve the MonoFlex by +4.65% improvement on AP@0.7 for car detection, reaching 18.54% AP@0.7, which ranks the 1st place among all monocular based methods on KITTI test leaderboard. The code and pretrained models will be released at https://github.com/yanglei18/Mix-Teaching.
Towards Adaptive Unknown Authentication for Universal Domain Adaptation by Classifier Paradox
Authors: Yunyun Wang, Yao Liu, Songcan Chen
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Universal domain adaptation (UniDA) is a general unsupervised domain adaptation setting, which addresses both domain and label shifts in adaptation. Its main challenge lies in how to identify target samples in unshared or unknown classes. Previous methods commonly strive to depict sample "confidence" along with a threshold for rejecting unknowns, and align feature distributions of shared classes across domains. However, it is still hard to pre-specify a "confidence" criterion and threshold which are adaptive to various real tasks, and a mis-prediction of unknowns further incurs misalignment of features in shared classes. In this paper, we propose a new UniDA method with adaptive Unknown Authentication by Classifier Paradox (UACP), considering that samples with paradoxical predictions are probably unknowns belonging to none of the source classes. In UACP, a composite classifier is jointly designed with two types of predictors. That is, a multi-class (MC) predictor classifies samples to one of the multiple source classes, while a binary one-vs-all (OVA) predictor further verifies the prediction by MC predictor. Samples with verification failure or paradox are identified as unknowns. Further, instead of feature alignment for shared classes, implicit domain alignment is conducted in output space such that samples across domains share the same decision boundary, though with feature discrepancy. Empirical results validate UACP under both open-set and universal UDA settings.
Keyword: scaling
Out of Distribution Detection via Neural Network Anchoring
Abstract
Our goal in this paper is to exploit heteroscedastic temperature scaling as a calibration strategy for out of distribution (OOD) detection. Heteroscedasticity here refers to the fact that the optimal temperature parameter for each sample can be different, as opposed to conventional approaches that use the same value for the entire distribution. To enable this, we propose a new training strategy called anchoring that can estimate appropriate temperature values for each sample, leading to state-of-the-art OOD detection performance across several benchmarks. Using NTK theory, we show that this temperature function estimate is closely linked to the epistemic uncertainty of the classifier, which explains its behavior. In contrast to some of the best-performing OOD detection approaches, our method does not require exposure to additional outlier datasets, custom calibration objectives, or model ensembling. Through empirical studies with different OOD detection settings -- far OOD, near OOD, and semantically coherent OOD - we establish a highly effective OOD detection approach. Code and models can be accessed here -- https://github.com/rushilanirudh/AMP
Strong Anonymity for Mesh Messaging
Authors: Neil Perry, Bruce Spang, Saba Eskandarian, Dan Boneh
Abstract
Messaging systems built on mesh networks consisting of smartphones communicating over Bluetooth have been used by protesters around the world after governments have disrupted Internet connectivity. Unfortunately, existing systems have been shown to be insecure; most concerningly by not adequately hiding metadata. This is further complicated by the fact that wireless communication such as Bluetooth is inherently a broadcasting medium. In this paper, we present a new threat model that captures the security requirements of protesters in this setting. We then provide a solution that satisfies the required security properties, hides all relevant metadata, scales to moderately sized protests, and supports group messaging. This is achieved by broadcasting all messages in a way that limits the overhead of duplicate messages, ensuring that ciphertexts do not leak metadata, and limiting what can be learned by observing user behavior. We also build a model of our system and numerically evaluate it to support our claims and analyze how many users it supports. Finally, we discuss further extensions that remove potential bottlenecks in scaling and support substantially more users.
Distributed-Memory Parallel Contig Generation for De Novo Long-Read Genome Assembly
Authors: Giulia Guidi, Gabriel Raulet, Daniel Rokhsar, Leonid Oliker, Katherine Yelick, Aydin Buluc
Subjects: Distributed, Parallel, and Cluster Computing (cs.DC); Genomics (q-bio.GN)
Abstract
De novo genome assembly, i.e., rebuilding the sequence of an unknown genome from redundant and erroneous short sequences, is a key but computationally intensive step in many genomics pipelines. The exponential growth of genomic data is increasing the computational demand and requires scalable, high-performance approaches. In this work, we present a novel distributed-memory algorithm that, from a string graph representation of the genome and using sparse matrices, generates the contig set, i.e., overlapping sequences that form a map representing a region of a chromosome. Using matrix abstraction, we mask branches in the string graph and compute the connected component to group genomic sequences that belong to the same linear chain (i.e., contig). Then, we perform multiway number partitioning to minimize the load imbalance in local assembly, i.e., concatenation of sequences from a given contig. Based on the assignment obtained by partitioning, we compute the induce subgraph function to redistribute sequences between processes, resulting in a set of local sparse matrices. Finally, we traverse each matrix using depth-first search to concatenate sequences. Our algorithm shows good scaling with parallel efficiency up to 80% on 128 nodes, resulting in uniform genome coverage and showing promising results in terms of assembly quality. Our contig generation algorithm localizes the assembly process to significantly reduce the amount of computation spent on this step. Our work is a step forward for efficient de novo long read assembly of large genomes in a distributed memory.
Accelerating Polarization via Alphabet Extension
Authors: Iwan Duursma, Ryan Gabrys, Venkatesan Guruswami, Ting-Chun Lin, Hsin-Po Wang
Abstract
Polarization is an unprecedented coding technique in that it not only achieves channel capacity, but also does so at a faster speed of convergence than any other technique. This speed is measured by the "scaling exponent" and its importance is three-fold. Firstly, estimating the scaling exponent is challenging and demands a deeper understanding of the dynamics of communication channels. Secondly, scaling exponents serve as a benchmark for different variants of polar codes that helps us select the proper variant for real-life applications. Thirdly, the need to optimize for the scaling exponent sheds light on how to reinforce the design of polar code. In this paper, we generalize the binary erasure channel (BEC), the simplest communication channel and the protagonist of many polar code studies, to the "tetrahedral erasure channel" (TEC). We then invoke Mori--Tanaka's $2 \times 2$ matrix over GF$(4)$ to construct polar codes over TEC. Our main contribution is showing that the dynamic of TECs converges to an almost--one-parameter family of channels, which then leads to an upper bound of $3.328$ on the scaling exponent. This is the first non-binary matrix whose scaling exponent is upper-bounded. It also polarizes BEC faster than all known binary matrices up to $23 \times 23$ in size. Our result indicates that expanding the alphabet is a more effective and practical alternative to enlarging the matrix in order to achieve faster polarization.
Abstract
Standard gradient descent algorithms applied to sequences of tasks are known to produce catastrophic forgetting in deep neural networks. When trained on a new task in a sequence, the model updates its parameters on the current task, forgetting past knowledge. This article explores scenarios where we scale the number of tasks in a finite environment. Those scenarios are composed of a long sequence of tasks with reoccurring data. We show that in such setting, stochastic gradient descent can learn, progress, and converge to a solution that according to existing literature needs a continual learning algorithm. In other words, we show that the model performs knowledge retention and accumulation without specific memorization mechanisms. We propose a new experimentation framework, SCoLe (Scaling Continual Learning), to study the knowledge retention and accumulation of algorithms in potentially infinite sequences of tasks. To explore this setting, we performed a large number of experiments on sequences of 1,000 tasks to better understand this new family of settings. We also propose a slight modifications to the vanilla stochastic gradient descent to facilitate continual learning in this setting. The SCoLe framework represents a good simulation of practical training environments with reoccurring situations and allows the study of convergence behavior in long sequences. Our experiments show that previous results on short scenarios cannot always be extrapolated to longer scenarios.
Keyword: calibration
Out of Distribution Detection via Neural Network Anchoring
Abstract
Our goal in this paper is to exploit heteroscedastic temperature scaling as a calibration strategy for out of distribution (OOD) detection. Heteroscedasticity here refers to the fact that the optimal temperature parameter for each sample can be different, as opposed to conventional approaches that use the same value for the entire distribution. To enable this, we propose a new training strategy called anchoring that can estimate appropriate temperature values for each sample, leading to state-of-the-art OOD detection performance across several benchmarks. Using NTK theory, we show that this temperature function estimate is closely linked to the epistemic uncertainty of the classifier, which explains its behavior. In contrast to some of the best-performing OOD detection approaches, our method does not require exposure to additional outlier datasets, custom calibration objectives, or model ensembling. Through empirical studies with different OOD detection settings -- far OOD, near OOD, and semantically coherent OOD - we establish a highly effective OOD detection approach. Code and models can be accessed here -- https://github.com/rushilanirudh/AMP
Abstract
This paper describes the systems developed by the HCCL team for the NIST 2021 speaker recognition evaluation (NIST SRE21).We first explore various state-of-the-art speaker embedding extractors combined with a novel circle loss to obtain discriminative deep speaker embeddings. Considering that cross-channel and cross-linguistic speaker recognition are the key challenges of SRE21, we introduce several techniques to reduce the cross-domain mismatch. Specifically, Codec and speech enhancement are directly applied to the raw speech to eliminate the codecs and the environment noise mismatch. We denote the methods that work directly on speech to eliminate the relatively explicit mismatches collectively as data adaptation methods. Experiments show that data adaption methods achieve 15\% improvements over our baseline. Furthermore, some popular back-ends domain adaptation algorithms are deployed on speaker embeddings to alleviate speaker performance degradation caused by the implicit mismatch. Score calibration is a major failure for us in SRE21. The reason is that score calibration with too many parameters easily lead to overfitting problems.
Consistency is the key to further mitigating catastrophic forgetting in continual learning
Abstract
Deep neural networks struggle to continually learn multiple sequential tasks due to catastrophic forgetting of previously learned tasks. Rehearsal-based methods which explicitly store previous task samples in the buffer and interleave them with the current task samples have proven to be the most effective in mitigating forgetting. However, Experience Replay (ER) does not perform well under low-buffer regimes and longer task sequences as its performance is commensurate with the buffer size. Consistency in predictions of soft-targets can assist ER in preserving information pertaining to previous tasks better as soft-targets capture the rich similarity structure of the data. Therefore, we examine the role of consistency regularization in ER framework under various continual learning scenarios. We also propose to cast consistency regularization as a self-supervised pretext task thereby enabling the use of a wide variety of self-supervised learning methods as regularizers. While simultaneously enhancing model calibration and robustness to natural corruptions, regularizing consistency in predictions results in lesser forgetting across all continual learning scenarios. Among the different families of regularizers, we find that stricter consistency constraints preserve previous task information in ER better.
Parallel Bayesian Optimization of Agent-based Transportation Simulation
Authors: Kiran Chhatre, Sidney Feygin, Colin Sheppard, Rashid Waraich
Subjects: Machine Learning (cs.LG); Multiagent Systems (cs.MA)
Abstract
MATSim (Multi-Agent Transport Simulation Toolkit) is an open source large-scale agent-based transportation planning project applied to various areas like road transport, public transport, freight transport, regional evacuation, etc. BEAM (Behavior, Energy, Autonomy, and Mobility) framework extends MATSim to enable powerful and scalable analysis of urban transportation systems. The agents from the BEAM simulation exhibit 'mode choice' behavior based on multinomial logit model. In our study, we consider eight mode choices viz. bike, car, walk, ride hail, driving to transit, walking to transit, ride hail to transit, and ride hail pooling. The 'alternative specific constants' for each mode choice are critical hyperparameters in a configuration file related to a particular scenario under experimentation. We use the 'Urbansim-10k' BEAM scenario (with 10,000 population size) for all our experiments. Since these hyperparameters affect the simulation in complex ways, manual calibration methods are time consuming. We present a parallel Bayesian optimization method with early stopping rule to achieve fast convergence for the given multi-in-multi-out problem to its optimal configurations. Our model is based on an open source HpBandSter package. This approach combines hierarchy of several 1D Kernel Density Estimators (KDE) with a cheap evaluator (Hyperband, a single multidimensional KDE). Our model has also incorporated extrapolation based early stopping rule. With our model, we could achieve a 25% L1 norm for a large-scale BEAM simulation in fully autonomous manner. To the best of our knowledge, our work is the first of its kind applied to large-scale multi-agent transportation simulations. This work can be useful for surrogate modeling of scenarios with very large populations.
Keyword: out of distribution detection
There is no result
Keyword: out-of-distribution detection
There is no result
Keyword: expected calibration error
There is no result
Keyword: overconfident
There is no result
Keyword: overconfidence
There is no result
Keyword: confidence
Mix-Teaching: A Simple, Unified and Effective Semi-Supervised Learning Framework for Monocular 3D Object Detection
Towards Adaptive Unknown Authentication for Universal Domain Adaptation by Classifier Paradox
Keyword: scaling
Out of Distribution Detection via Neural Network Anchoring
Strong Anonymity for Mesh Messaging
Distributed-Memory Parallel Contig Generation for De Novo Long-Read Genome Assembly
Accelerating Polarization via Alphabet Extension
Scaling the Number of Tasks in Continual Learning
Keyword: calibration
Out of Distribution Detection via Neural Network Anchoring
The HCCL System for the NIST SRE21
Consistency is the key to further mitigating catastrophic forgetting in continual learning
Parallel Bayesian Optimization of Agent-based Transportation Simulation