Abstract
Unsupervised Domain Adaptation (UDA) for object detection aims to adapt a model trained on a source domain to detect instances from a new target domain for which annotations are not available. Different from traditional approaches, we propose ConfMix, the first method that introduces a sample mixing strategy based on region-level detection confidence for adaptive object detector learning. We mix the local region of the target sample that corresponds to the most confident pseudo detections with a source image, and apply an additional consistency loss term to gradually adapt towards the target data distribution. In order to robustly define a confidence score for a region, we exploit the confidence score per pseudo detection that accounts for both the detector-dependent confidence and the bounding box uncertainty. Moreover, we propose a novel pseudo labelling scheme that progressively filters the pseudo target detections using the confidence metric that varies from a loose to strict manner along the training. We perform extensive experiments with three datasets, achieving state-of-the-art performance in two of them and approaching the supervised target model performance in the other. Code is available at: https://github.com/giuliomattolin/ConfMix.
XC: Exploring Quantitative Use Cases for Explanations in 3D Object Detection
Authors: Sunsheng Gu, Vahdat Abdelzad, Krzysztof Czarnecki
Abstract
Explainable AI (XAI) methods are frequently applied to obtain qualitative insights about deep models' predictions. However, such insights need to be interpreted by a human observer to be useful. In this paper, we aim to use explanations directly to make decisions without human observers. We adopt two gradient-based explanation methods, Integrated Gradients (IG) and backprop, for the task of 3D object detection. Then, we propose a set of quantitative measures, named Explanation Concentration (XC) scores, that can be used for downstream tasks. These scores quantify the concentration of attributions within the boundaries of detected objects. We evaluate the effectiveness of XC scores via the task of distinguishing true positive (TP) and false positive (FP) detected objects in the KITTI and Waymo datasets. The results demonstrate an improvement of more than 100\% on both datasets compared to other heuristics such as random guesses and the number of LiDAR points in the bounding box, raising confidence in XC's potential for application in more use cases. Our results also indicate that computationally expensive XAI methods like IG may not be more valuable when used quantitatively compare to simpler methods.
Large Language Models Can Self-Improve
Authors: Jiaxin Huang, Shixiang Shane Gu, Le Hou, Yuexin Wu, Xuezhi Wang, Hongkun Yu, Jiawei Han
Abstract
Large Language Models (LLMs) have achieved excellent performances in various tasks. However, fine-tuning an LLM requires extensive supervision. Human, on the other hand, may improve their reasoning abilities by self-thinking without external inputs. In this work, we demonstrate that an LLM is also capable of self-improving with only unlabeled datasets. We use a pre-trained LLM to generate "high-confidence" rationale-augmented answers for unlabeled questions using Chain-of-Thought prompting and self-consistency, and fine-tune the LLM using those self-generated solutions as target outputs. We show that our approach improves the general reasoning ability of a 540B-parameter LLM (74.4%->82.1% on GSM8K, 78.2%->83.0% on DROP, 90.0%->94.4% on OpenBookQA, and 63.4%->67.9% on ANLI-A3) and achieves state-of-the-art-level performance, without any ground truth label. We conduct ablation studies and show that fine-tuning on reasoning is critical for self-improvement.
Abstract
This paper proposes to perform online clustering by conducting twin contrastive learning (TCL) at the instance and cluster level. Specifically, we find that when the data is projected into a feature space with a dimensionality of the target cluster number, the rows and columns of its feature matrix correspond to the instance and cluster representation, respectively. Based on the observation, for a given dataset, the proposed TCL first constructs positive and negative pairs through data augmentations. Thereafter, in the row and column space of the feature matrix, instance- and cluster-level contrastive learning are respectively conducted by pulling together positive pairs while pushing apart the negatives. To alleviate the influence of intrinsic false-negative pairs and rectify cluster assignments, we adopt a confidence-based criterion to select pseudo-labels for boosting both the instance- and cluster-level contrastive learning. As a result, the clustering performance is further improved. Besides the elegant idea of twin contrastive learning, another advantage of TCL is that it could independently predict the cluster assignment for each instance, thus effortlessly fitting online scenarios. Extensive experiments on six widely-used image and text benchmarks demonstrate the effectiveness of TCL. The code will be released on GitHub.
Diffusion Visual Counterfactual Explanations
Authors: Maximilian Augustin, Valentyn Boreiko, Francesco Croce, Matthias Hein
Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Abstract
Visual Counterfactual Explanations (VCEs) are an important tool to understand the decisions of an image classifier. They are 'small' but 'realistic' semantic changes of the image changing the classifier decision. Current approaches for the generation of VCEs are restricted to adversarially robust models and often contain non-realistic artefacts, or are limited to image classification problems with few classes. In this paper, we overcome this by generating Diffusion Visual Counterfactual Explanations (DVCEs) for arbitrary ImageNet classifiers via a diffusion process. Two modifications to the diffusion process are key for our DVCEs: first, an adaptive parameterization, whose hyperparameters generalize across images and models, together with distance regularization and late start of the diffusion process, allow us to generate images with minimal semantic changes to the original ones but different classification. Second, our cone regularization via an adversarially robust model ensures that the diffusion process does not converge to trivial non-semantic changes, but instead produces realistic images of the target class which achieve high confidence by the classifier.
RT-MOT: Confidence-Aware Real-Time Scheduling Framework for Multi-Object Tracking Tasks
Abstract
Different from existing MOT (Multi-Object Tracking) techniques that usually aim at improving tracking accuracy and average FPS, real-time systems such as autonomous vehicles necessitate new requirements of MOT under limited computing resources: (R1) guarantee of timely execution and (R2) high tracking accuracy. In this paper, we propose RT-MOT, a novel system design for multiple MOT tasks, which addresses R1 and R2. Focusing on multiple choices of a workload pair of detection and association, which are two main components of the tracking-by-detection approach for MOT, we tailor a measure of object confidence for RT-MOT and develop how to estimate the measure for the next frame of each MOT task. By utilizing the estimation, we make it possible to predict tracking accuracy variation according to different workload pairs to be applied to the next frame of an MOT task. Next, we develop a novel confidence-aware real-time scheduling framework, which offers an offline timing guarantee for a set of MOT tasks based on non-preemptive fixed-priority scheduling with the smallest workload pair. At run-time, the framework checks the feasibility of a priority-inversion associated with a larger workload pair, which does not compromise the timing guarantee of every task, and then chooses a feasible scenario that yields the largest tracking accuracy improvement based on the proposed prediction. Our experiment results demonstrate that RT-MOT significantly improves overall tracking accuracy by up to 1.5x, compared to existing popular tracking-by-detection approaches, while guaranteeing timely execution of all MOT tasks.
Keyword: scaling
SMaLL-100: Introducing Shallow Multilingual Machine Translation Model for Low-Resource Languages
Abstract
In recent years, multilingual machine translation models have achieved promising performance on low-resource language pairs by sharing information between similar languages, thus enabling zero-shot translation. To overcome the "curse of multilinguality", these models often opt for scaling up the number of parameters, which makes their use in resource-constrained environments challenging. We introduce SMaLL-100, a distilled version of the M2M-100 (12B) model, a massively multilingual machine translation model covering 100 languages. We train SMaLL-100 with uniform sampling across all language pairs and therefore focus on preserving the performance of low-resource languages. We evaluate SMaLL-100 on different low-resource benchmarks: FLORES-101, Tatoeba, and TICO-19 and demonstrate that it outperforms previous massively multilingual models of comparable sizes (200-600M) while improving inference latency and memory usage. Additionally, our model achieves comparable results to M2M-100 (1.2B), while being 3.6x smaller and 4.3x faster at inference. Code and pre-trained models: https://github.com/alirezamshi/small100
Graphically Structured Diffusion Models
Authors: Christian Weilbach, William Harvey, Frank Wood
Subjects: Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE); Programming Languages (cs.PL)
Abstract
We introduce a framework for automatically defining and learning deep generative models with problem-specific structure. We tackle problem domains that are more traditionally solved by algorithms such as sorting, constraint satisfaction for Sudoku, and matrix factorization. Concretely, we train diffusion models with an architecture tailored to the problem specification. This problem specification should contain a graphical model describing relationships between variables, and often benefits from explicit representation of subcomputations. Permutation invariances can also be exploited. Across a diverse set of experiments we improve the scaling relationship between problem dimension and our model's performance, in terms of both training time and final accuracy.
LiteVL: Efficient Video-Language Learning with Enhanced Spatial-Temporal Modeling
Authors: Dongsheng Chen, Chaofan Tao, Lu Hou, Lifeng Shang, Xin Jiang, Qun Liu
Subjects: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
Abstract
Recent large-scale video-language pre-trained models have shown appealing performance on various downstream tasks. However, the pre-training process is computationally expensive due to the requirement of millions of video-text pairs and the redundant data structure of each video. To mitigate these problems, we propose LiteVL, which adapts a pre-trained image-language model BLIP into a video-text model directly on downstream tasks, without heavy pre-training. To enhance the temporal modeling lacking in the image-language model, we propose to add temporal attention modules in the image encoder of BLIP with dynamic temporal scaling. Besides the model-wise adaptation, we also propose a non-parametric pooling mechanism to adaptively reweight the fine-grained video embedding conditioned on the text. Experimental results on text-video retrieval and video question answering show that the proposed LiteVL even outperforms previous video-language pre-trained models by a clear margin, though without any video-language pre-training.
Energy Efficiency of Massive Random Access in MIMO Quasi-Static Rayleigh Fading Channels with Finite Blocklength
Authors: Junyuan Gao, Yongpeng Wu, Shuo Shao, Wei Yang, H. Vincent Poor
Abstract
This paper considers the massive random access problem in MIMO quasi-static Rayleigh fading channels. Specifically, we derive achievability and converse bounds on the minimum energy-per-bit required for each active user to transmit $J$ bits with blocklength $n$ and power $P$ under a per-user probability of error (PUPE) constraint, in the cases with and without \emph{a priori} channel state information at the receiver (CSIR and no-CSI). In the case of no-CSI, we consider both the settings with and without knowing the number $K_a$ of active users. The achievability bounds rely on the design of the ``good region''. Numerical evaluation shows the gap between achievability and converse bounds is less than $2.5$ dB in the CSIR case and less than $4$ dB in the no-CSI case in most considered regimes. When the distribution of $K_a$ is known, the performance gap between the cases with and without knowing the value of $K_a$ is small. For example, in the setup with blocklength $n=1000$, payload $J=100$, error requirement $\epsilon=0.001$, and $L=128$ receive antennas, compared to the case with known $K_a$, the extra required energy-per-bit when $K_a$ is unknown and distributed as $K_a\sim\text{Binom}(K,0.4)$ is less than $0.3$ dB on the converse side and $1.1$ dB on the achievability side. The spectral efficiency grows approximately linearly with $L$ in the CSIR case, whereas the growth rate decreases with no-CSI. Moreover, we study the performance of a pilot-assisted scheme, which is suboptimal especially when $K_a$ is large. Building on non-asymptotic results, when all users are active and $J=\Theta(1)$, we obtain scaling laws as follows: when $L=\Theta \left(n^2\right)$ and $P=\Theta\left(\frac{1}{n^2}\right)$, one can reliably serve $K=\mathcal{O}(n^2)$ users with no-CSI; under mild conditions with CSIR, the PUPE requirement is satisfied if and only if $\frac{nL\ln KP}{K}=\Omega\left(1\right)$.
Efficient Dataset Distillation Using Random Feature Approximation
Authors: Noel Loo, Ramin Hasani, Alexander Amini, Daniela Rus
Abstract
Dataset distillation compresses large datasets into smaller synthetic coresets which retain performance with the aim of reducing the storage and computational burden of processing the entire dataset. Today's best-performing algorithm, \textit{Kernel Inducing Points} (KIP), which makes use of the correspondence between infinite-width neural networks and kernel-ridge regression, is prohibitively slow due to the exact computation of the neural tangent kernel matrix, scaling $O(|S|^2)$, with $|S|$ being the coreset size. To improve this, we propose a novel algorithm that uses a random feature approximation (RFA) of the Neural Network Gaussian Process (NNGP) kernel, which reduces the kernel matrix computation to $O(|S|)$. Our algorithm provides at least a 100-fold speedup over KIP and can run on a single GPU. Our new method, termed an RFA Distillation (RFAD), performs competitively with KIP and other dataset condensation algorithms in accuracy over a range of large-scale datasets, both in kernel regression and finite-width network training. We demonstrate the effectiveness of our approach on tasks involving model interpretability and privacy preservation.
Keyword: calibration
Encoding nonlinear and unsteady aerodynamics of limit cycle oscillations using nonlinear sparse Bayesian learning
Authors: Rimple Sandhu, Brandon Robinson, Mohammad Khalil, Chris L. Pettit, Dominique Poirel, Abhijit Sarkar
Subjects: Computational Engineering, Finance, and Science (cs.CE)
Abstract
This paper investigates the applicability of a recently-proposed nonlinear sparse Bayesian learning (NSBL) algorithm to identify and estimate the complex aerodynamics of limit cycle oscillations. NSBL provides a semi-analytical framework for determining the data-optimal sparse model nested within a (potentially) over-parameterized model. This is particularly relevant to nonlinear dynamical systems where modelling approaches involve the use of physics-based and data-driven components. In such cases, the data-driven components, where analytical descriptions of the physical processes are not readily available, are often prone to overfitting, meaning that the empirical aspects of these models will often involve the calibration of an unnecessarily large number of parameters. While it may be possible to fit the data well, this can become an issue when using these models for predictions in regimes that are different from those where the data was recorded. In view of this, it is desirable to not only calibrate the model parameters, but also to identify the optimal compromise between data-fit and model complexity. In this paper, this is achieved for an aeroelastic system where the structural dynamics are well-known and described by a differential equation model, coupled with a semi-empirical aerodynamic model for laminar separation flutter resulting in low-amplitude limit cycle oscillations. For the purpose of illustrating the benefit of the algorithm, in this paper, we use synthetic data to demonstrate the ability of the algorithm to correctly identify the optimal model and model parameters, given a known data-generating model. The synthetic data are generated from a forward simulation of a known differential equation model with parameters selected so as to mimic the dynamics observed in wind-tunnel experiments.
Adaptive re-calibration of channel-wise features for Adversarial Audio Classification
Abstract
DeepFake Audio, unlike DeepFake images and videos, has been relatively less explored from detection perspective, and the solutions which exist for the synthetic speech classification either use complex networks or dont generalize to different varieties of synthetic speech obtained using different generative and optimization-based methods. Through this work, we propose a channel-wise recalibration of features using attention feature fusion for synthetic speech detection and compare its performance against different detection methods including End2End models and Resnet-based models on synthetic speech generated using Text to Speech and Vocoder systems like WaveNet, WaveRNN, Tactotron, and WaveGlow. We also experiment with Squeeze Excitation (SE) blocks in our Resnet models and found that the combination was able to get better performance. In addition to the analysis, we also demonstrate that the combination of Linear frequency cepstral coefficients (LFCC) and Mel Frequency cepstral coefficients (MFCC) using the attentional feature fusion technique creates better input features representations which can help even simpler models generalize well on synthetic speech classification tasks. Our models (Resnet based using feature fusion) trained on Fake or Real (FoR) dataset and were able to achieve 95% test accuracy with the FoR data, and an average of 90% accuracy with samples we generated using different generative models after adapting this framework.
DIICAN: Dual Time-scale State-Coupled Co-estimation of SOC, SOH and RUL for Lithium-Ion Batteries
Authors: Ningbo Cai, Yuwen Qin, Xin Chen, Kai Wu
Subjects: Systems and Control (eess.SY); Machine Learning (cs.LG); Data Analysis, Statistics and Probability (physics.data-an)
Abstract
Accurate co-estimations of battery states, such as state-of-charge (SOC), state-of-health (SOH,) and remaining useful life (RUL), are crucial to the battery management systems to assure safe and reliable management. Although the external properties of the battery charge with the aging degree, batteries' degradation mechanism shares similar evolving patterns. Since batteries are complicated chemical systems, these states are highly coupled with intricate electrochemical processes. A state-coupled co-estimation method named Deep Inter and Intra-Cycle Attention Network (DIICAN) is proposed in this paper to estimate SOC, SOH, and RUL, which organizes battery measurement data into the intra-cycle and inter-cycle time scales. And to extract degradation-related features automatically and adapt to practical working conditions, the convolutional neural network is applied. The state degradation attention unit is utilized to extract the battery state evolution pattern and evaluate the battery degradation degree. To account for the influence of battery aging on the SOC estimation, the battery degradation-related state is incorporated in the SOC estimation for capacity calibration. The DIICAN method is validated on the Oxford battery dataset. The experimental results show that the proposed method can achieve SOH and RUL co-estimation with high accuracy and effectively improve SOC estimation accuracy for the whole lifespan.
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
ConfMix: Unsupervised Domain Adaptation for Object Detection via Confidence-based Mixing
XC: Exploring Quantitative Use Cases for Explanations in 3D Object Detection
Large Language Models Can Self-Improve
Twin Contrastive Learning for Online Clustering
Diffusion Visual Counterfactual Explanations
RT-MOT: Confidence-Aware Real-Time Scheduling Framework for Multi-Object Tracking Tasks
Keyword: scaling
SMaLL-100: Introducing Shallow Multilingual Machine Translation Model for Low-Resource Languages
Graphically Structured Diffusion Models
LiteVL: Efficient Video-Language Learning with Enhanced Spatial-Temporal Modeling
Energy Efficiency of Massive Random Access in MIMO Quasi-Static Rayleigh Fading Channels with Finite Blocklength
Efficient Dataset Distillation Using Random Feature Approximation
Keyword: calibration
Encoding nonlinear and unsteady aerodynamics of limit cycle oscillations using nonlinear sparse Bayesian learning
Adaptive re-calibration of channel-wise features for Adversarial Audio Classification
DIICAN: Dual Time-scale State-Coupled Co-estimation of SOC, SOH and RUL for Lithium-Ion Batteries