Brandeis Ph.D. seminars are hosted at most once every two weeks. Any Ph.D. can sign up to give a talk, and all Brandeis CS and CL Ph.D. students are welcome to attend. Free pizza is provided. To sign up, email Ryan Marcus. Seminars are currently held on Fridays at 2pm in Volen 101.
Previous and upcoming seminars:
We present a method to extend the idea of sprite coding to videos containing a wide variety of naturally occurring background motion, which could potentially be incorporated into existing and future video standards. The existing MPEG-4 part 2 standard, now almost 20 years old, provides the ability to store objects in separate layers, and includes a sprite mode where the background layer is generated by cropping a still image based on frame-wide global motion parameters, but videos containing more general background motion cannot be effectively encoded with sprite mode. We propose a perceptually motivated lossy compression algorithm, where oscillatory background motion can be compactly encoded. Our model achieves a low bit rate by referencing a time-invariant representation of the optical flow with only a few added parameters per frame. At very low bit rates, our technique can provide dynamic backgrounds at a visual quality that may not be achievable by traditional methods which are known to produce unacceptable blocking and ringing artifacts.
Integrating machine learning into the internals of database management systems requires significant feature engineering, a human effort-intensive process to determine the best way to represent the pieces of information that are relevant to a task. In addition to being labor intensive, the process of hand-engineering features must generally be repeated for each data management task, and may make assumptions about the underlying database that are not universally true. We introduce flexible operator embeddings, a deep learning technique for automatically transforming query operators into feature vectors that are useful for a multiple data management tasks and is custom-tailored to the underlying database. Our approach works by taking advantage of an operator's context, resulting in a neural network that quickly transforms sparse representations of query operators into dense, information-rich feature vectors. Experimentally, we show that our flexible operator embeddings perform well across a number of data management tasks, using both synthetic and real-world datasets.
We examined whether signals from inexpensive, wearable brainwave sensors could be used to identify the STEM learning task in which a student was engaged. Twelve subjects completed four different STEM learning tasks – two entailing passive learning (watching a video or reading), and two entailing active learning (solving problems based on the passive learning). There were two mathematics tasks (one active and one passive) and two Python programming tasks (one active, one passive). Subjects were fitted with wearable brainwave sensors that captured cortical oscillations from four scalp electrodes, and transformed the signals from each electrode into five distinct frequency bands. This yielded 10 samples per second within each frequency band and from each electrode. We then trained ensemble-based machine learning algorithms (boosting and bagging of decision tree learners) to classify various features of tasks and subjects from a single sample of brainwave activity. We explored several different types of training/testing regimes, and our results suggest that within a single session, brain activity patterns for each of these four types of learning are substantially different, but that the patterns do not generalize well between sessions. Importantly, the brainwave patterns differ greatly between individuals, which suggests that applications using such devices will need to rely on personalization to achieve high accuracy. The project is a first step toward developing apps that could use individualized EEG feedback to help subjects develop learning strategies that optimize their learning experience.
Distributed data management systems often operate on “elastic” clusters that can scale up or down on demand. These systems face numerous challenges, including data fragmentation, replication, and cluster sizing. Unfortunately, these challenges have traditionally been treated independently, leaving administrators with little insight on how the interplay of these decisions affects query performance. This paper introduces NashDB, an adaptive data distribution framework that relies on an economic model to automatically balance the supply and demand of data fragments, replicas, and cluster nodes. NashDB adapts its decisions to query priorities and shifting workloads, while avoiding underutilized cluster nodes and redun dant replicas. This paper introduces and evaluates NashDB’s model, as well as a suite of optimization techniques designed to efficiently identify data distribution schemes that match workload demands and transition the system to this new scheme with minimum data transfer overhead. Experimentally, we show that NashDB is often Pareto dominant compared to other solutions.
Applications need to analyze the past state of their data to provide auditing and other forms of fact checking. Retrospective snapshot systems that support computations over data store snapshots, allow applications using simple data stores like Berkeley DB or SQLite, to provide past state analysis in a convenient way. Current snapshot systems however, offer no satisfactory support for computations that analyze multiple snapshots. We have developed a Retrospective Query Language (RQL), a simple declarative extension to SQL that allows to specify and run multi-snapshot computations conveniently in a snapshot system, using a small number of simple mechanisms defined in terms of relational constructs familiar to programmers. We describe RQL mechanisms, explain how they translate into SQL computations in a snapshot system, and show how to express a number of common analysis patterns with illustrative examples. We also describe how we implemented RQL in a simple way utilizing SQLite UDF framework in a Berkeley DB data store using Retro page-level incremental snapshot system. Multi-snapshot computations running over page-level incremental snapshots bring up interesting performance issues that have not been studied before. We present the first study defining a performance envelope for multi-snapshot computations over page-level incremental snapshots.
The advent of wearable consumer-grade brainwave sensors opens the possibility of building educational technology that can provide reliable feedback about the focus and attention of a student who is engaged in a learning activity. In this paper, we demonstrate the practicality of developing a simple web-based application that exploits EEG data to monitor reading effectiveness personalized for individual readers. Our tool uses a variant of k-means classification on the relative power of the five standard bands (alpha, beta, gamma, delta, theta) for each of four electrodes on the Muse wearable brainwave sensor. We demonstrate that after 30 minutes of training, our relatively simple approach is able to successfully distinguish between brain signals produced when the subject engages in reading versus when they are relaxing. The accuracy of classification varied across the 10 subjects from 55% to 85% with a mean of 71%. The standard approach to recognize relaxation is to look for strong alpha and/or theta signals and it is reasonably effective but is most associated with closed eye relaxation and it does not allow for personalization. Our k-means classification approach provides a personalized classifier which distinguishes open eye relaxation from reading and has the potential to detect a wide variety of different cognitive states.
In this talk I will introduce a framework in which virtual agents learn to enact complex temporal-spatial actions by observing humans. My framework processes motion capture data of human subjects performing actions, and uses sequential modeling on qualitative spatial features to learn representations of these actions. Using reinforcement learning, these observed sequences are used to guide a simulated agent to perform novel actions.
CNNs are poised to become integral parts of many critical systems. Despite their robustness to natural variations, image pixel values can be manipulated, via small, carefully crafted, imperceptible perturbations, to cause a model to misclassify images. We present an algorithm to process an image so that classification accuracy is significantly preserved in the presence of such adversarial manipulations. Image classifiers tend to be robust to natural noise, and adversarial attacks tend to be agnostic to object location. These observations motivate our strategy, which leverages model robustness to defend against adversarial perturbations by forcing the image to match natural image statistics. Our algorithm locally corrupts the image by redistributing pixel values via a process we term pixel deflection. A subsequent wavelet-based denoising operation softens this corruption, as well as some of the adversarial changes. We demonstrate experimentally that the combination of these techniques enables the effective recovery of the true class, against a variety of robust attacks. Our results compare favorably with current state-of-the-art defenses, without requiring retraining or modifying the CNN.
We propose a linguistic prediction game with competitive and cooperative variants, and a model of game players based on finite state automata. We present a complexity metric for these automata, and study the coevolutionary dynamics of complexity growth in a variety of multi-species simulations. We present quantitative results using this complexity metric and analyze the causes of varying rates of complexity growth across different types of interactions. We find that while both purely competitive and purely cooperative coevolution are able to drive complexity growth above the rate of genetic drift, mixed systems with both competitive and cooperative interactions achieve significantly higher evolved complexity.
Workload management for cloud databases deals with the tasks of resource provisioning, query placement, and query scheduling in a manner that meets the application’s performance goals while minimizing the cost of using cloud resources. Existing solutions have approached these three challenges in isolation while aiming to optimize a single performance metric. In this paper, we introduce WiSeDB, a learning-based framework for generating holistic workload management solutions customized to application-defined performance goals and workload characteristics. Our approach relies on supervised learning to train cost-effective decision tree models for guiding query placement, scheduling, and resource provisioning decisions. Applications can use these models for both batch and online scheduling of incoming workloads. A unique feature of our system is that it can adapt its offline model to stricter/looser performance goals with minimal re-training. This allows us to present to the application alternative workload management strategies that address the typical performance vs. cost trade-off of cloud services. Experimental results show that our approach has very low training overhead while offering low cost strategies for a variety of performance metrics and workload characteristics.
Diagnosis of a clinical condition is a challenging task, which often requires significant medical investigation. Previous work related to diagnostic inferencing problems mostly consider multivariate observational data (e.g. physiological signals, lab tests etc.). In contrast, we explore the problem using free-text medical notes recorded in an electronic health record (EHR). Complex tasks like these can benefit from structured knowledge bases, but those are not scalable. We instead exploit raw text from Wikipedia as a knowledge source. Memory networks have been demonstrated to be effective in tasks which require comprehension of free-form text. They use the final iteration of the learned representation to predict probable classes. We introduce condensed memory neural networks (C-MemNNs), a novel model with iterative condensation of memory representations that preserves the hierarchy of features in the memory. Experiments on the MIMIC-III dataset show that the proposed model outperforms other variants of memory networks to predict the most probable diagnoses given a complex clinical scenario.
In this paper, we describe a system for generating three-dimensional visual simulations of natural language motion expressions. We use a rich formal model of events and their participants to generate simulations that satisfy the minimal constraints entailed by the associated utterance, relying on semantic knowledge of physical objects and motion events. This paper outlines technical considerations and discusses implementing the aforementioned semantic models into such a system.
We demonstrate a self-organizing, multi-agent system to generate approximate solutions to the route assignment problem for a large number of vehicles across many origins and destinations. Our algorithm produces a set of mixed strategies over the set of paths through the network, which are suitable for use by autonomous vehicles in the absence of centralized control or coordination. Our approach combines ideas from co-evolutionary dynamics in which many species coordinate and compete for efficient navigation, and ideas from swarm intelligence in which many simple agents self-organize into successful behavior using limited between-agent communication. Experiments demonstrate a marked improvement of both individual and total travel times as compared to greedy uncoordinated strategies, and we analyze the differences in outcomes for various routes as the simulation progresses.
Inferring implicit discourse relations in natural language text is the most difficult subtask in discourse parsing. Many neural network models have been proposed to tackle this problem. However, the comparison for this task is not unified, so we could hardly draw clear conclusions about the effectiveness of various architectures. Here, we propose neural network models that are based on feedforward and long-short term memory architecture and systematically study the effects of varying structures. To our surprise, the best-configured feedforward architecture outperforms LSTM-based model in most cases despite thorough tuning. Further, we compare our best feedforward system with competitive convolutional and recurrent networks and find that feedforward can actually be more effective. For the first time for this task, we compile and publish outputs from previous neural and non-neural systems to establish the standard for further comparison.