Abhishek Mukherji

MobileMiner: mining your frequent patterns on your phone

Smartphones can collect considerable context data about the user, ranging from apps used to places visited. Frequent user patterns discovered from longitudinal, multi-modal context data could help personalize and improve overall user experience. Our long term goal is to develop novel middleware and algorithms to efficiently mine user behavior patterns entirely on the phone by utilizing idle processor cycles. Mining patterns on the mobile device provides better privacy guarantees to users, and reduces dependency on cloud connectivity. As an important step in this direction, we develop a novel general-purpose service called MobileMiner that runs on the phone and discovers frequent co-occurrence patterns indicating which context events frequently occur together. Using longitudinal context data collected from 106 users over 1--3 months, we show that MobileMiner efficiently generates patterns using limited phone resources. Further, we find interesting behavior patterns for individual users and across users, ranging from calling patterns to place visitation patterns. Finally, we show how our co-occurrence patterns can be used by developers to improve the phone UI for launching apps or calling contacts.

PARAS: a parameter space framework for online association mining

Association rule mining is known to be computationally intensive, yet real-time decision-making applications are increasingly intolerant to delays. In this paper, we introduce the parameter space model, called PARAS. PARAS enables efficient rule mining by compactly maintaining the final rulesets. The PARAS model is based on the notion of stable region abstractions that form the coarse granularity ruleset space. Based on new insights on the redundancy relationships among rules, PARAS establishes a surprisingly compact representation of complex redundancy relationships while enabling efficient redundancy resolution at query-time. Besides the classical rule mining requests, the PARAS model supports three novel classes of exploratory queries. Using the proposed PSpace index, these exploratory query classes can all be answered with near real-time responsiveness. Our experimental evaluation using several benchmark datasets demonstrates that PARAS achieves 2 to 5 orders of magnitude improvement over state-of-the-art approaches in online association rule mining.

Adding intelligence to your mobile device via on-device sequential pattern mining

The next revolution in mobile user experience is predicted to be a smart device that can adapt to its users lifestyle and surroundings to become a proactive personal assistant. We introduce the idea of Mobile Sequence Mining (MSM) engine that automatically learns phone usage sequential patterns over the rich context data captured within the device. The learned patterns can then enable variety of applications including proactive assistance for a variety of use cases. Unlike existing cloud-based intelligence services (e.g., GoogleNow) that rely on internet access and may compromise privacy, MSM provides device intelligence by leveraging mined longitudinal patterns while preserving privacy via on-device mining. MSM is generic and can provide sequential patterns and predictions over multiple data streams, also allowing individual mobile applications to stream their own private data to mine sequential patterns. In our preliminary tests by deploying MSM on 3 user devices, it mines frequent sequential patterns within 8 minutes over 7-53 days of longitudinal user context data including location, app usage and call logs spanning 137-312 unique contexts. We conclude the paper by enumerating future research challenges for mobile sequence mining.

PARAS: interactive parameter space exploration for association rule mining

We demonstrate our PARAS technology for supporting interactive association mining at near real-time speeds. Key technical innovations of PARAS, in particular, stable region abstractions and rule redundancy management supporting novel parameter space-centric exploratory queries will be showcased. The audience will be able to interactively explore the parameter space view of rules. They will experience near real-time speeds achieved by PARAS for operations, such as comparing rule sets mined using different parameter values, that would otherwise take hours of computation and much manual investigation. Overall, we will demonstrate that the PARAS system provides a rich experience to data analysts through parameter tuning recommendations while significantly reducing the trial-and-error interactions.

FireStream: Sensor Stream Processing for Monitoring Fire Spread

This demonstration presents FireStream, a sensor stream processing system which provides services for run-time detection, monitoring and visualization of fire spread in intelligent buildings that can be of great benefit to first responders. Our system can effectively handle large heterogeneous sensor streams using shared window execution and dynamic participant handling to yield a high-ary MJoin solution.

Did you take a break today? Detecting playing foosball using your smartwatch

Prolonged working hours are a primary cause of stress, work related injuries (e.g, RSIs), and work-life imbalance in employees at a workplace. As reported by some studies, taking timely breaks from continuous work not only reduces stress and exhaustion but also improves productivity, employee bonding, and camaraderie. Our goal is to build a system that automatically detects breaks thereby assisting in maintaining healthy work-break balance. In this paper, we focus on detecting foosball breaks of employees at a workplace using a smartwatch. We selected foosball as it is one of the most commonly played games at many workplaces in the United States. Since playing foosball involves wrist and hand movement, a wrist-worn device (e.g., a smartwatch), due to its position, has a clear advantage over a smartphone for detecting foosball activity. Our evaluation using data collected from real workplace shows that we can identify with more than 95% accuracy whether a person is playing foosball or not. We also show that we can determine how long a foosball session lasted with an error of less than 3% in the best case.

FIRE: interactive visual support for parameter space-driven rule mining

While significant strides have been made on efficient association rule mining, the usability of mining systems woefully lags behind. In particular, the usability of rule mining systems is limited by the lack of support for interactive exploration of the relationships among rule results produced with various parameter settings. Based on a novel parameter space-driven approach, our proposed Framework for Interactive Rule Exploration (FIRE) addresses the usability shortcoming. FIRE features innovative visual displays and effective interactions that enable analysts to conduct rule exploration at the speed of thought. Particularly, the parameter space view (PSpace) displays the distribution of rules produced for diverse parameter settings. This not only facilitates user parameter selection but also empowers analysts to understand rule relationships in the parameter space context. Our user study with 22 subjects establishes the usability and effectiveness of the proposed features and interactions of FIRE using benchmark datasets. Overall, this research encompasses significant contributions at the intersection of data mining, knowledge management and visual analytics.

SPIRE: supporting parameter-driven interactive rule mining and exploration

We demonstrate our SPIRE technology for supporting interactive mining of both positive and negative rules at the speed of thought. It is often misleading to learn only about positive rules, yet extremely revealing to find strongly supported negative rules. Key technical contributions of SPIRE including region-wise abstractions of rules, positive-negative rule relationship analysis, rule redundancy management and rule visualization supporting novel exploratory queries will be showcased. The audience can interactively explore complex rule relationships in a visual manner, such as comparing negative rules with their positive counterparts, that would otherwise take prohibitive time. Overall, our SPIRE system provides data analysts with rich insights into rules and rule relationships while significantly reducing manual effort and time investment required.

SNIF TOOL: sniffing for patterns in continuous streams

Continuous time-series sequence matching, specifically, matching a numeric live stream against a set of redefined pattern sequences, is critical for domains ranging from fire spread tracking to network traffic monitoring. While several algorithms exist for similarity matching of static time-series data, matching continuous data poses new, largely unsolved challenges including online real-time processing requirements and system resource limitations for handling infinite streams. In this work, we propose a novel live stream matching framework, called n-Snippet Indices Framework (in short, SNIF), to tackle these challenges. SNIF employs snippets as the basic unit for matching streaming time-series. The insight is to perform the matching at two levels of granularity: bag matching of subsets of snippets of the live stream against prefixes of the patterns, and order checking for maintaining successive candidate snippet bag matches. We design a two-level index structure, called SNIF index, which supports these two modes of matching. We propose a family of online two-level prefix matching algorithms that trade off between result accuracy and response time. The effectiveness of SNIF to detect patterns has been thoroughly tested through experiments using real datasets from the domains of fire monitoring and sensor motes. In this paper, we also present a study of SNIFs performance, accuracy and tolerance to noise compared against those of the state-of-the-art Continuous Query with Prediction (CQP) approach.

WatchUDrive: Differentiating drivers and passengers using smartwatches

Personalization and automation in future smart vehicles hinge on accurately identifying the driver and passengers in the vehicle. Traditional approaches either require additional infrastructure or impose assumptions about how users interact with their smartphones. The recent proliferation of commercial smartwatches enables new opportunities to solve this problem due to the fixed position of the watch on the wrist. We use this observation to motivate WatchUDrive, our smartwatch-based application for identifying whether the wearer is the driver or a passenger in a vehicle. We evaluate two smartwatch sensing modalities for driver vs. passenger differentiation: the accelerometer and the camera. Using 40 in-vehicle episodes collected from 8 users and 8 different vehicles, we show that the accelerometer yields 90% accuracy within 10 seconds, whereas the camera only yields 62% accuracy within 110 seconds.