Roadside LiDAR (Light Detection and Ranging) sensors promise safer and faster traffic management and vehicular operations. However, occlusion and small view angles are significant challenges to widespread use of roadside LiDARs. We consider fusing data from multiple LiDARs at a traffic intersection to better estimate traffic parameters than one can estimate from a single LiDAR. The key challenge is to calibrate multiple LiDARs both in time and space. The problem is more complex when heterogeneous sensors differ in resolution and are positioned arbitrarily on a traffic intersection.

Retrieval-augmented generation (RAG) systems combine the strengths of language generation and information retrieval to power many real-world applications like chatbots. Use of RAG for understanding of videos is appealing but there are two critical limitations. One-time, upfront conversion of all content in large corpus of videos into text descriptions entails high processing times. Also, not all information in the rich video data is typically captured in the text descriptions. Since user queries are not known apriori, developing a system for video to text conversion and interactive querying of video data is challenging.

JPEG remains one of the most widespread lossy image coding methods. However, the non-differentiable nature of JPEG restricts the application in deep learning pipelines. Several differentiable approximations of JPEG have recently been proposed to address this issue. This paper conducts a comprehensive review of existing diff. JPEG approaches and identifies critical details that have been missed by previous methods. To this end, we propose a novel diff. JPEG approach, overcoming previous limitations. Our approach is differentiable w.r.t. the input image, the JPEG quality, the quantization tables, and the color conversion parameters. We evaluate the forward and backward performance of our diff. JPEG approach against existing methods. Additionally, extensive ablations are performed to evaluate crucial design choices. Our proposed diff. JPEG resembles the (non-diff.) reference implementation best, significantly surpassing the recent-best diff. approach by 3.47dB (PSNR) on average. For strong compression rates, we can even improve PSNR by 9.51dB. Strong adversarial attack results are yielded by our diff. JPEG, demonstrating the effective gradient approximation. Our code is available at https://github.com/necla-ml/Diff-JPEG.

Edge Intelligence has received attention in the recent times for its potential towards improving responsiveness, reducing the cost of data transmission, enhancing security and privacy, and enabling autonomous decisions by edge devices. However, edge devices lack the power and compute resources necessary to execute most Al models. In this paper, we present FactionFormer, a novel method to deploy resource-intensive deep-learning models, such as vision transformers (ViT), on resource-constrained edge devices. Our method is based on a key observation: edge devices are often deployed in settings where they encounter only a subset of the classes that the resource-intensive Al model is trained to classify, and this subset changes across deployments. Therefore, we automatically identify this subset as a faction, devise on-the fly a bespoke resource-efficient ViT called a modelette for the faction, and set up an efficient processing pipeline consisting of a modelette on the device, a wireless network such as 5G, and the resource-intensive ViT model on an edge server, all of which work collaboratively to do the inference. For several ViT models pre-trained on benchmark datasets, FactionFormer’s modelettes are up to 4× smaller than the corresponding baseline models in terms of the number of parameters, and they can infer up to 2.5× faster than the baseline setup where every input is processed by the resource-intensive ViT on the edge server. Our work is the first of its kind to propose a device-edge collaborative inference framework where bespoke deep learning models for the device are automatically devised on-the-fly for most frequently encountered subset of classes.

Roadside LiDAR (Light Detection and Ranging) sensors are recently being explored for intelligent transportation systems aiming at safer and faster traffic management and vehicular operations. A key challenge in such systems is to efficiently transfer massive point-cloud data from the roadside LiDAR devices to the edge connected through a 5G network for real-time processing. In this paper, we consider the problem of compressing roadside (i.e. static) LiDAR data in real-time that provides a unique condition unexplored by current methods. Existing point-cloud compression methods assume moving LiDARs (that are mounted on vehicles) and do not exploit spatial consistency across frames over time.

Meta learning algorithms for few-shot video recognition use complex, episodic training but they often fail to learn effective feature representations. In contrast, we propose a new and simpler few-shot video recognition method that does not use meta-learning, but its performance compares well with the best meta-learning proposals. Our new few-shot video classification pipeline consists of two distinct phases. In the pre-training phase, we learn a good video feature extraction network that generates a feature vector for each video. After a sparse sampling strategy selects frames from the video, we generate a video feature vector from the sampled frames. Our proposed video feature extractor network, which consists of an image feature extraction network followed by a new transformer encoder, is trained end-to-end by including a classifier head that uses cosine similarity layer instead of the traditional linear layer to classify a corpus of labeled video examples. Unlike prior work in meta learning, we do not use episodic training to learn the image feature vector. Also, unlike prior work that averages frame-level feature vectors into a single video feature vector, we combine individual frame-level feature vectors by using a new Transformer encoder that explicitly captures the key, temporal properties in the sequence of sampled frames. End-to-end training of the video feature extractor ensures that the proposed Transformer encoder captures important temporal properties in the video, while the cosine similarity layer explicitly reduces the intra-class variance of videos that belong to the same class. Next, in the few-shot adaptation phase, we use the learned video feature extractor to train a new video classifier by using the few available examples from novel classes. Results on SSV2-100 and Kinetics-100 benchmarks show that our proposed few-shot video classifier outperforms the meta-learning-based methods and achieves the best state-of-the-art accuracy. We also show that our method can easily discern between actions and their inverse (for example, picking something up vs. putting something down), while prior art, which averages image feature vectors, is unable to do so.

Design of multitasking deep learning models has mostly focused on improving the accuracy of the constituent tasks, but the challenges of efficiently deploying such models in a device-edge collaborative setup (that is common in 5G deployments) has not been investigated. Towards this end, in this paper, we propose an approach called Chimera1for training (done Offline) and deployment (done Online) of multitasking deep learning models that are splittable across the device and edge. In the offline phase, we train our multi-tasking setup such that features from a pre-trained model for one of the tasks (called the Primary task) are extracted and task-specific sub-models are trained to generate the other (Secondary) tasks' outputs through a knowledge distillation like training strategy to mimic the outputs of pre-trained models for the tasks. The task-specific sub-models are designed to be significantly lightweight than the original pre-trained models for the Secondary tasks. Once the sub-models are trained, during deployment, for given deployment context, characterized by the configurations, we search for the optimal (in terms of both model performance and cost) deployment strategy for the generated multitasking model, through finding one or multiple suitable layer(s) for splitting the model, so that inference workloads are distributed between the device and the edge server and the inference is done in a collaborative manner. Extensive experiments on benchmark computer vision tasks demonstrate that Chimera generates splittable multitasking models that are at least ~ 3 x parameter efficient than the existing such models, and the end-to-end device-edge collaborative inference becomes ~ 1.35 x faster with our choice of context-aware splitting decisions.

Identification of people with elevated body temperature can reduce or dramatically slow down the spread of infectious diseases like COVID-19. We present a novel fever-screening system, F3S, that uses edge machine learning techniques to accurately measure core body temperatures of multiple individuals in a free-flow setting. F3S performs real-time sensor fusion of visual camera with thermal camera data streams to detect elevated body temperature, and it has several unique features: (a) visual and thermal streams represent very different modalities, and we dynamically associate semantically-equivalent regions across visual and thermal frames by using a new, dynamic alignment technique that analyzes content and context in real-time, (b) we track people through occlusions, identify the eye (inner canthus), forehead, face and head regions where possible, and provide an accurate temperature reading by using a prioritized refinement algorithm, and (c) we robustly detect elevated body temperature even in the presence of personal protective equipment like masks, or sunglasses or hats, all of which can be affected by hot weather and lead to spurious temperature readings. F3S has been deployed at over a dozen large commercial establishments, providing contact-less, free-flow, real-time fever screening for thousands of employees and customers in indoors and outdoor settings.

Administrators of most user-facing systems depend on periodic log data to get an idea of the health and status of production applications. Logs report information, which is crucial to diagnose the root cause of complex problems. In this paper, we present a real-time log analysis system called LogLens that automates the process of anomaly detection from logs with no (or minimal) target system knowledge and user specification. In LogLens, we employ unsupervised machine learning based techniques to discover patterns in application logs, and then leverage these patterns along with the real-time log parsing for designing advanced log analytics applications. Compared to the existing systems which are primarily limited to log indexing and search capabilities, LogLens presents an extensible system for supporting both stateless and stateful log analysis applications. Currently, LogLens is running at the core of a commercial log analysis solution handling millions of logs generated from the large-scale industrial environments and reported up to 12096x man-hours reduction in troubleshooting operational problems compared to the manual approach.

Modern engineering incorporates smart technologies in all aspects of our lives. Smart technologies are generating terabytes of log messages every day to report their status. It is crucial to analyze these log messages and present usable information (e.g. patterns) to administrators, so that they can manage and monitor these technologies. Patterns minimally represent large groups of log messages and enable the administrators to do further analysis, such as anomaly detection and event prediction. Although patterns exist commonly in automated log messages, recognizing them in massive set of log messages from heterogeneous sources without any prior information is a significant undertaking. We propose a method, named LogMine, that extracts high quality patterns for a given set of log messages. Our method is fast, memory efficient, accurate, and scalable. LogMine is implemented in map-reduce framework for distributed platforms to process millions of log messages in seconds. LogMine is a robust method that works for heterogeneous log messages generated in a wide variety of systems. Our method exploits algorithmic techniques to minimize the computational overhead based on the fact that log messages are always automatically generated. We evaluate the performance of LogMine on massive sets of log messages generated in industrial applications. LogMine has successfully generated patterns which are as good as the patterns generated by exact and unscalable method, while achieving a 500× speedup. Finally, we describe three applications of the patterns generated by LogMine in monitoring large scale industrial systems.

The performance and capacity characteristics of flash storage make it attractive to use as a cache. Recency-based cache replacement policies rely on an in-memory full index, typically a B-tree or a hash table, that maps each object to its recency information. Even though the recency information itself may take very little space, the full index for a cache holding N keys requires at least log N bits per key. This metadata overhead is undesirably high when used for very large flash-based caches, such as key-value stores with billions of objects. To solve this problem, we propose a new RAM-frugal cache replacement policy that approximates the least-recently-used (LRU) policy. It uses two in-memory Bloom sub-filters (TBF) for maintaining the recency information and leverages an on-flash key-value store to cache objects. TBF requires only one byte of RAM per cached object, making it suitable for implementing very large flash-based caches. We evaluate TBF through simulation on traces from several block stores and key-value stores, as well as evaluate it using the Yahoo! Cloud Serving Benchmark in a real system implementation. Evaluation results show that TBF achieves cache hit rate and operations per second comparable to those of LRU in spite of its much smaller memory requirements.