In this paper, we propose a framework for efficient Source-Free Domain Adaptation (SFDA) in the context of time-series, focusing on enhancing both parameter efficiency and data-sample utilization. Our approach introduces an improved paradigm for source-model preparation and target-side adaptation, aiming to enhance training efficiency during target adaptation. Specifically, we reparameterize the source model's weights in a Tucker-style decomposed manner, factorizing the model into a compact form during the source model preparation phase. During target-side adaptation, only a subset of these decomposed factors is fine-tuned, leading to significant improvements in training efficiency. We demonstrate using PAC Bayesian analysis that this selective fine-tuning strategy implicitly regularizes the adaptation process by constraining the model's learning capacity. Furthermore, this re-parameterization reduces the overall model size and enhances inference efficiency, making the approach particularly well suited for resource-constrained devices. Additionally, we demonstrate that our framework is compatible with various SFDA methods and achieves significant computational efficiency, reducing the number of fine-tuned parameters and inference overhead in terms of MACs by over 90\% while maintaining model performance.

Data-free Knowledge Distillation (DFKD) has gained popularity recently, with the fundamental idea of carrying out knowledge transfer from a Teacher neural network to a Student neural network in the absence of training data. However, in the Adversarial DFKD framework, the student network's accuracy, suffers due to the non-stationary distribution of the pseudo-samples under multiple generator updates. To this end, at every generator update, we aim to maintain the student's performance on previously encountered examples while acquiring knowledge from samples of the current distribution. Thus, we propose a meta-learning inspired framework by treating the task of Knowledge-Acquisition (learning from newly generated samples) and Knowledge-Retention (retaining knowledge on previously met samples) as meta-train and meta-test, respectively. Hence, we dub our method as Learning to Retain while Acquiring. Moreover, we identify an implicit aligning factor between the Knowledge-Retention and Knowledge-Acquisition tasks indicating that the proposed student update strategy enforces a common gradient direction for both tasks, alleviating interference between the two objectives. Finally, we support our hypothesis by exhibiting extensive evaluation and comparison of our method with prior arts on multiple datasets.

This paper looks at semi-supervised learning (SSL) for image-based text recognition. One of the most popular SSL approaches is pseudo-labeling (PL). PL approaches assign labels to unlabeled data before re-training the model with a combination of labeled and pseudo-labeled data. However, PL methods are severely degraded by noise and are prone to over-fitting to noisy labels, due to the inclusion of erroneous high confidence pseudo-labels generated from poorly calibrated models, thus, rendering threshold-based selection ineffective. Moreover, the combinatorial complexity of the hypothesis space and the error accumulation due to multiple incorrect autoregressive steps posit pseudo-labeling challenging for sequence models. To this end, we propose a pseudo-label generation and an uncertainty-based data selection framework for semi-supervised text recognition. We first use Beam-Search inference to yield highly probable hypotheses to assign pseudo-labels to the unlabelled examples. Then we adopt an ensemble of models, sampled by applying dropout, to obtain a robust estimate of the uncertainty associated with the prediction, considering both the character-level and word-level predictive distribution to select good quality pseudo-labels. Extensive experiments on several benchmark handwriting and scene-text datasets show that our method outperforms the baseline approaches and the previous state-of- the-art semi-supervised text-recognition methods.

This paper investigates a novel learning-based camera motion characterization scheme along with its application to the video stabilization problem. The proposed characterization scheme represents the compressed domain block motion vectors (MVs) using polar angle and magnitude histograms. Discriminative features from these two histograms are extracted and fed to a supervised learning-based hierarchical classifier for recognizing the six camera motion patterns. A comparative analysis with an existing scheme is carried out to support and validate the proposed characterization scheme. The proposed scheme works at the frame level by classifying the inter-frame camera motion patterns. This scheme is extended to classify the video segments and a novel application to video stabilization is investigated. An experimental analysis of a number of test sequences captured using a handheld video camera shows that by characterizing the smooth and jittery motions, selective video stabilization could be carried out only on those video segments that have been degraded. This approach of selective video stabilization saves considerable amount of computational time compared with running the stabilization algorithm on the entire video sequence, as proposed in the literature. The proposed strategy of using a classification scheme prior to applying the video stabilization routine offers a new paradigm to the conventional video stabilization problem. Experimental validation carried out using exhaustive search motion estimation obtained block MVs, and H.264/Advanced Video Coding-obtained MVs shows that by using the idea of selective video stabilization, up to 62% reduction in processing time can be achieved compared with video stabilization approaches wherein the entire video sequence is processed.

In this paper, we present the architecture of a highly configurable multi-input multi-output (MIMO) orthogonal frequency division multiple access (OFDMA) platform. The platform is designed to support experimentation with various communication algorithms, thus allowing an intimate understanding of the performance of complex algorithms under real-life constraints. The hardware used is the wireless open access research platform (WARP) which facilitates rapid prototyping utilizing the FPGA and multi radio interfaces available.

Signature files have been studied extensively, as an access method for textual databases. Many approaches have been proposed for searching signatures files efficiently. However, different methods make different assumptions and use different performance measures, making it difficult to compare their performance. In this paper, we study three basic methods proposed in the literature, namely, the indexed descriptor file, the two-level superimposed coding scheme, and the partitioned signature file approach. The contribution of this paper is two-fold. First, we present a uniform analytical performance model so that the methods can be compared fairly and consistently. The analysis shows that the two-level superimposed coding scheme, if stored in a transposed file, has the best performance. Second, we extend the two-level superimposed coding method into a multilevel superimposed coding method, we obtain the optimal number of levels for the multilevel method and show that for databases with reasonable size the optimal value is much larger than 2, which is assumed in the two-level method. The accuracy of the analytical formula is demonstrated by simulation.>