Skip to main content

Karim Ali

Followers

0

Following

1

Public Views

Interests

Uploads

Papers by Karim Ali

Toward an Analytical Performance Model to Select between GPU and CPU Execution

2019 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)

Automating the device selection in heterogeneous computing platforms requires the modelling of pe... more Automating the device selection in heterogeneous computing platforms requires the modelling of performance both on CPUs and on accelerators. This work argues for the use of a hybrid analytical performance modelling approach is a practical way to build fast and efficient methods to select an appropriate target for a given computation kernel. The target selection problem has been addressed in the literature, however there has been a strong emphasis on building empirical models with machine learning techniques. We argue that the applicability of such solutions is often limited in production systems. This paper focus on the issue of building a selector to decide if an OpenMP loop nest should be executed in a CPU or in a GPU. To this end, it offers a comprehensive comparison evaluation of the difference in GPU kernel performance on devices of multiple generations of architectures. The idea is to underscore the need for accurate analytical performance models and to provide insights in the evolution of GPU accelerators. This work also highlights a drawback of existing approaches to modelling GPU performance-accurate modelling of memory coalescing characteristics. To that end, we examine a novel application of an inter-thread difference analysis that can further improve analytical models. Finally, this work presents an initial study of an Open Multi-Processing (OpenMP) runtime framework for target-offloading target selection.

Compiler-driven performance

The compiler-driven performance workshop will consist of the presentation of reports on research ... more The compiler-driven performance workshop will consist of the presentation of reports on research progress at various academic and industrial sites across Canada and in the United States. Topics to be discussed in the workshop will include, but are not limited to: • innovative compiler analysis, transformation, and optimization techniques • languages, compilers, and optimization techniques for multicore processors and other parallel architectures • analysis and compilation techniques for dynamic programming languages • compiling for streaming or heterogeneous hardware • dynamic compilation for high-performance and real-time environments • compilation and optimization for scripting languages • compilation techniques for reducing power • program safety • whole system optimization and analysis • tools and infrastructure for compiler research

Supplementary material from "Memory reactivation in rat medial prefrontal cortex occurs in a subtype of cortical UP state during slow-wave sleep

Interaction between hippocampal sharp-wave ripples (SWRs) and UP states, possibly by coordinated ... more Interaction between hippocampal sharp-wave ripples (SWRs) and UP states, possibly by coordinated reactivation of memory traces, is conjectured to play an important role in memory consolidation. Recently, it was reported that SWRs were differentiated into multiple subtypes. However, whether cortical UP states can also be classified into subtypes is not known. Here, we analysed neural ensemble activity from the medial prefrontal cortex from rats trained to run a spatial sequence-memory task. Application of the hidden Markov model (HMM) with three states to epochs of UP-DOWN oscillations identified DOWN states and two subtypes of UP state (UP-1 and UP-2). The two UP subtypes were distinguished by differences in duration, with UP-1 having a longer duration than UP-2, as well as differences in the speed of population vector (PV) decorrelation, with UP-1 decorrelating more slowly than UP-2. Reactivation of recent memory sequences predominantly occurred in UP-2. Short-duration reactivating UP states were dominated by UP-2 whereas long-duration ones exhibit transitions from UP-1 to UP-2. Thus, recent memory reactivation, if it occurred within long-duration UP states, typically was preceded by a period of slow PV evolution not related to recent experience, and which we speculate may be related to previously encoded information. If that is the case, then the transition from UP-1 to UP-2 subtypes may help gradual integration of recent experience with pre-existing cortical memories by interleaving the two in the same UP state.'This article is part of the Theo Murphy meeting issue 'Memory reactivation: replaying events past, present and future'.

CogniCrypt: Supporting developers in using cryptography

2017 32nd IEEE/ACM International Conference on Automated Software Engineering (ASE), 2017

Previous research suggests that developers often struggle using low-level cryptographic APIs and,... more Previous research suggests that developers often struggle using low-level cryptographic APIs and, as a result, produce insecure code. When asked, developers desire, among other things, more tool support to help them use such APIs. In this paper, we present CogniCrypt, a tool that supports developers with the use of cryptographic APIs. CogniCrypt assists the developer in two ways. First, for a number of common cryptographic tasks, CogniCrypt generates code that implements the respective task in a secure manner. Currently, CogniCrypt supports tasks such as data encryption, communication over secure channels, and long-term archiving. Second, CogniCrypt continuously runs static analyses in the background to ensure a secure integration of the generated code into the developer's workspace. This video demo showcases the main features of CogniCrypt: youtube.com/watch?v=JUq5mRHfAWY.

SRAM-PUF-Based Entities Authentication Scheme for Resource-Constrained IoT Devices

IEEE Internet of Things Journal, 2021

With the development of the cloud-based Internet of Things (IoT), people and things can request s... more With the development of the cloud-based Internet of Things (IoT), people and things can request services, access data, or control actuators located thousands of miles away. The entity authentication of the remotely accessed devices is an essential part of the security systems. In this vein, Physical Unclonable Functions (PUFs) are a hot research topic, especially for generating random, stable, and tamper-resistant fingerprints. This paper proposes a lightweight, robust SRAM-PUF based entity authentication scheme to guarantee that the accessed end devices are trustable. The proposed scheme uses Challenge-Response Pairs (CRPs) represented by reordered memory addresses as challenges and the corresponding SRAM cells' startup values as responses. The experimental results show that our scheme can efficiently authenticate resources-constrained IoT devices with a low computation overhead and small memory capacity. Furthermore, we analyze the SRAM-PUF by testing the PUF output under different environmental conditions, including temperature and magnetic field, in addition to exploring the effect of writing different values to the SRAM cells on the stability of their startup values.

Assessment of Sleep, K-Complexes, and Sleep Spindles in a T21 Light-Dark Cycle

Frontiers in Neuroscience, 2020

Circadian rhythm misalignment has a deleterious impact on the brain and the body. In rats, exposu... more Circadian rhythm misalignment has a deleterious impact on the brain and the body. In rats, exposure to a 21-hour day length impairs hippocampal dependent memory. Sleep, and particularly K-complexes and sleep spindles in the cortex, have been hypothesized to be involved in memory consolidation. Altered K-complexes, sleep spindles, or interaction between the cortex and hippocampus could be a mechanism for the memory consolidation failure but has yet to be assessed in any circadian misalignment paradigm. In the current study, continuous local field potential recordings from five rats were used to assess the changes in aspects of behavior and sleep, including wheel running activity, quiet wakefulness, motionless sleep, slow wave sleep, REM sleep, K-complexes and sleep spindles, in rats exposed to six consecutive days of a T21 light-dark cycle (L9:D12). Except for a temporal redistribution of sleep and activity during the T21, there were no changes in period, or total amount for any aspect of sleep or activity. These data suggest that the memory impairment elicited from 6 days of T21 exposure is likely not due to changes in sleep architecture. It remains possible that hippocampal plasticity is affected by experiencing light when subjective circadian phase is calling for dark. However, if there is a reduction in hippocampal plasticity, changes in sleep appear not to be driving this effect.

Lifting the curse of stringly-typed code

How often do JavaScript programmers embed structured languages into strings literals? We conduct ... more How often do JavaScript programmers embed structured languages into strings literals? We conduct an empirical investigating mining nearly 500 thousand JavaScript source files from almost ten thousand repositories from GitHub. We parsed each string literal with seven separate common grammars, and found the most common data type that is hidden within the confines of string literals. To reduce the overuse of strings for structured data types, we present a simple static program analyzer that finds embedded languages and warns the developer, providing an optional fix.

Debugging Static Analysis

IEEE Transactions on Software Engineering, 2019

To detect and x bugs and security vulnerabilities, software companies use static analysis as part... more To detect and x bugs and security vulnerabilities, software companies use static analysis as part of the development process. However, static analysis code itself is also prone to bugs. To ensure a consistent level of precision, as analyzed programs grow more complex, a static analysis has to handle more code constructs, frameworks, and libraries that the programs use. While more complex analyses are written and used in production systems every day, the cost of debugging and xing them also increases tremendously. To better understand the di culties of debugging static analyses, we surveyed 115 static analysis writers. From their responses, we extracted the core requirements to build a debugger for static analysis, which revolve around two main issues: (1) abstracting from two code bases at the same time (the analysis code and the analyzed code) and (2) tracking the analysis internal state throughout both code bases. Most current debugging tools that our survey participants use lack the capabilities to address both issues. Focusing on those requirements, we introduce V , a debugging environment for static data-ow analysis that is integrated in the Eclipse development environment. V features graph visualizations that enable users to view the state of a data-ow analysis and its intermediate results at any time. Special breakpoints in V help users step through the analysis code and the analyzed simultaneously. To evaluate the usefulness of V , we have conducted a user study on 20 static analysis writers. Using V helped our sample of analysis writers identify 25% and x 50% more errors in the analysis code compared to using the standard Eclipse debugging environment. CCS CONCEPTS •Software and its engineering → Software testing and debugging; •Theory of computation → Program analysis; •Humancentered computing → Empirical studies in visualization;

Energy- and Area-Efficient Spin–Orbit Torque Nonvolatile Flip-Flop for Power Gating Architecture

IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2018

This paper proposes two enhanced architectures for spin-orbit torque nonvolatile flip-flop (SOT-N... more This paper proposes two enhanced architectures for spin-orbit torque nonvolatile flip-flop (SOT-NVFF), namely, parallel programming scheme (pSOT-NVFF) and series programming scheme (sSOT-NVFF). At given SOT magnetic tunnel junction (SOT-MTJ) technology, the pSOT-NVFF achieves 29% lower backup energy with nearly 50% shorter backup delay compared with previous parallel architectures and the sSOT-NVFF offers at least 42% lower backup energy with 52% higher speed compared with others. Further analysis of the two proposed architectures showed a preferred window of operation for each architecture as a function of heavy metal electrode resistance (R H M). In particular, sSOT-NVFF is preferred, from speed and energy perspectives, for SOT-MTJs with small R H M (i.e., hundreds of to tens of k), while pSOT-NVFF is preferred for SOT-MTJs with large R H M in the range of hundreds of k. Further study indicates that our NVFF is robust, stable, and functional even with process variations. Index Terms-Nonvolatile flip-flop (NVFF), parallel programming, power gating (PG), series programming, spin-orbit torque magnetic tunnel junction (SOT-MTJ).

Context-, flow-, and field-sensitive data-flow analysis using synchronized Pushdown systems

Proceedings of the ACM on Programming Languages, 2019

Precise static analyses are context-, field- and flow-sensitive. Context- and field-sensitivity a... more Precise static analyses are context-, field- and flow-sensitive. Context- and field-sensitivity are both expressible as context-free language (CFL) reachability problems. Solving both CFL problems along the same data-flow path is undecidable, which is why most flow-sensitive data-flow analyses over-approximate field-sensitivity through k -limited access-path, or through access graphs. Unfortunately, as our experience and this paper show, both representations do not scale very well when used to analyze programs with recursive data structures. Any single CFL-reachability problem is efficiently solvable, by means of a pushdown system. This work thus introduces the concept of synchronized pushdown systems (SPDS). SPDS encode both procedure calls/returns and field stores/loads as separate but “synchronized” CFL reachability problems. An SPDS solves both individual problems precisely, and approximation occurs only in corner cases that are apparently rare in practice: at statements where b...

Towards secure integration of cryptographic software

2015 ACM International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software (Onward!), 2015

While cryptography is now readily available to everyone and can, provably, protect private inform... more While cryptography is now readily available to everyone and can, provably, protect private information from attackers, we still frequently hear about major data leakages, many of which are due to improper use of cryptographic mechanisms. The problem is that many application developers are not cryptographic experts. Even though high-quality cryptographic APIs are widely available, programmers often select the wrong algorithms or misuse APIs due to a lack of understanding. Such issues arise with both simple operations such as encryption as well as with complex secure communication protocols such as SSL. In this paper, we provide a long-term solution that helps application developers integrate cryptographic components correctly and securely by bridging the gap between cryptographers and application developers. Our solution consists of a software product line (with an underlying feature model) that automatically identifies the correct cryptographic algorithms to use, based on the developer's answers to high-level questions in non-expert terminology. Each feature (i.e., cryptographic algorithm) maps into corresponding Java code and a usage protocol describing API restrictions. By composing the user's selected features, we automatically synthesize a secure code blueprint and a usage protocol that corresponds to the selected usage scenario. Since the developer may change the application code over time, we use the usage protocols to statically analyze the program and ensure that the correct use of the API is not violated over time.

Constructing Call Graphs of Scala Programs

Lecture Notes in Computer Science, 2014

Cons ist en t * Comple te * We ll Docum en ted * Easy to * EC OOP * A rt ifact * A EC As Scala ga... more Cons ist en t * Comple te * We ll Docum en ted * Easy to * EC OOP * A rt ifact * A EC As Scala gains popularity, there is growing interest in programming tools for it. Such tools often require call graphs. However, call graph construction algorithms in the literature do not handle Scala features, such as traits and abstract type members. Applying existing call graph construction algorithms to the JVM bytecodes generated by the Scala compiler produces very imprecise results due to type information being lost during compilation. We adapt existing call graph construction algorithms, Name-Based Resolution (RA) and Rapid Type Analysis (RTA), for Scala, and present a formalization based on Featherweight Scala. We evaluate our algorithms on a collection of Scala programs. Our results show that careful handling of complex Scala constructs greatly helps precision and that our most precise analysis generates call graphs with 1.1-3.7 times fewer nodes and 1.5-18.7 times fewer edges than a bytecode-based RTA analysis. 2 Background Algorithms for call graph construction [11] have been studied extensively in the context of object-oriented programming languages such as Java [10, 17], C++ [6] and Self [1], of functional programming languages such as Scheme [24] and ML [12], and of scripting languages such as JavaScript [25]. Roughly speaking, most call graph construction algorithms can be classified as being either type-based or flow-based [7, 13, 14, 17, 18]. The former class of algorithms uses only local information given by static types to determine possible call targets, whereas the latter analyzes the program's data flow. We focus on type-based algorithms, so we will briefly review some important type-based call graph construction algorithms for object-oriented languages upon which our work is based. In the exposition of these algorithms, we use a constraint notation that is equivalent to that of [27], but that explicitly represents call graph edges using a relation ' →' between call sites and methods. Name-Based Resolution (RA). The main challenge in constructing call graphs of object-oriented programs is in approximating the behavior of dynamically dispatched (virtual) method calls. Early work (see, e.g., [26]) simply assumed that a virtual call e.m(• • •) can invoke any method with the same name m. This approach can be captured using the following constraints: main ∈ R RA main c → M M ∈ R RA reachable call c : e.m(.. .) occurs in method M method M has name m M ∈ R c → M RA call Intuitively, rule RA main reads "the main method is reachable" by including it in the set R of reachable methods. Rule RA call states that "if a method is reachable, and a call site c : e.m(.. .) occurs in its body, then every method with name m is reachable from c." Finally, rule RA reachable states that any method M reachable from a call site c is contained in the set R of reachable methods. Class Hierarchy Analysis (CHA). Obviously, Name-Based Resolution can become very imprecise if a class hierarchy contains unrelated methods that happen to have the same name. Class Hierarchy Analysis [9] improves upon name-based resolution by using the static type of the receiver expression of a method call in combination with class hierarchy information to determine what methods may be invoked from a call site. Following the notation of [27], we use StaticType(e) to denote the static type of an expression e, and StaticLookup(C, m) to denote the method definition that is invoked when method m is invoked on an object with run-time type C. Using these definitions, CHA is defined as follows: main ∈ R CHA main c → M M ∈ R CHA reachable call c : e.m(.. .) occurs in method M C ∈ SubTypes(StaticType(e)) StaticLookup(C, m) = M M ∈ R c → M CHA call

An Evaluation of Call Graph Construction Algorithms for Scala

As Scala gains popularity, there is growing interest in programming tools for it. Such tools ofte... more As Scala gains popularity, there is growing interest in programming tools for it. Such tools often require call graphs. However, call graph construction algorithms in the literature do not handle Scala features, such as traits and abstract type members. Applying existing call graph construction algorithms to the JVM bytecodes generated by the Scala compiler produces very imprecise results due to type information being lost during compilation. We adapt existing call graph construction algorithms, Name-Based Resolution (RA) and Rapid Type Analysis (RTA), for Scala, and present a formalization based on Featherweight Scala. We evaluate our algorithms on a collection of Scala programs. Our results show that careful handling of complex Scala constructs greatly helps precision and that our most precise analysis generates call graphs with 1.1-3.7 times fewer nodes and 1.5-18.7 times fewer edges than a bytecode-based RTA analysis.

The Separate Compilation Assumption

Call graphs are an essential requirement for almost all inter-procedural analyses. This motivated... more Call graphs are an essential requirement for almost all inter-procedural analyses. This motivated the development of many tools and frameworks to generate the call graph of a given program. However, the majority of these tools focus on generating the call graph of the whole program (i.e., both the application and the libraries that the application depends on). A popular compromise to the excessive cost of building a call graph for the whole program is to build an application-only call graph. To achieve this, all the effects of the library code and any calls that the library makes back into the application are usually ignored. This results in potential unsoundness in the generated call graph and therefore in analyses that use it. Additionally, the scope of the application classes to be analyzed by such an algorithm has been often arbitrarily defined.

Deconstructing the Deformable Parts Model: Do More with Less

The deformable parts model (DPM) [1] is a successful detection model that continues to achieve st... more The deformable parts model (DPM) [1] is a successful detection model that continues to achieve state-of-the-art results in object detection. While the model's success is attributed to the deformable parts, the proposed system has many other design choices such as the use of multi-resolution HOG features, the sampling of parts from high energy areas, the use of a mixture model, etc. Recently, there have been attempts at analyzing the contribution of these design choices more closely [2]. We delve further into this analysis, performing a study of the effects of restricting the deformation model, the use of single-resolution filters versus multi-resolution filters, and the application of energy "dropout" while sampling parts. Our results indicate that a better performance can be achieved with simpler cost-free deformation models.

FlowBoost — Appearance learning from sparsely annotated video

CVPR 2011, 2011

We propose a new learning method which exploits temporal consistency to successfully learn a comp... more We propose a new learning method which exploits temporal consistency to successfully learn a complex appearance model from a sparsely labeled training video. Our approach consists in iteratively improving an appearancebased model built with a Boosting procedure, and the reconstruction of trajectories corresponding to the motion of multiple targets. We demonstrate the efficiency of our procedure on pedestrian detection in videos and cell detection in microscopy image sequences. In both cases, our method is demonstrated to reduce the labeling requirement by one to two orders of magnitude. We show that in some instances, our method trained with sparse labels on a video sequence is able to outperform a standard learning procedure trained with the fully labeled sequence.

Ability of Triage Decision Rules for Rapid Electrocardiogram to Identify Patients With Suspected ST-Elevation Myocardial Infarction

Critical Pathways in Cardiology: A Journal of Evidence-Based Medicine, 2012

Background: The American College of Cardiology/American Heart Association guidelines for ST-eleva... more Background: The American College of Cardiology/American Heart Association guidelines for ST-elevation myocardial infarction state that an electrocardiogram (ECG) should be performed on patients with suspected acute coronary syndrome upon presentation to the emergency department (ED) within 10 minutes. Objective: To determine how previously published clinical criteria for obtaining an ECG at ED triage perform in a population of patients receiving emergency cardiac catheterization for suspected myocardial infarction. This rule was originally derived by Graff to identify clinical criteria for obtaining an ECG at triage to rapidly identify patients with acute myocardial infarction. The Graff rule was developed in a setting where lytic therapy was the primary reperfusion strategy. A modification proposed by Glickman adds several more criteria in an effort to capture additional patients. We hypothesized that the Graff rule would identify most patients for whom the cardiac catheterization laboratory (CCL) was activated and that the Glickman rule would capture the remaining patients. Methods: Three trained physician reviewers retrospectively applied the Graff decision rule to 430 consecutive patients from a database of emergency CCL activations by ED physicians. The Graff rule recommends that patients between the ages of 30 and 49 years received a rapid ECG if they complained of chest pain and those aged 50 years or older received a rapid ECG when they complained of chest pain, shortness of breath, palpitations, weakness, or syncope. The newly developed Glickman rule, which included nausea and vomiting in patients over the age of 80 years, was applied to the patients where the Graff rule was negative. The triage note or earliest medical contact documentation was used to determine whether the patient's complaints would have resulted in a rapid ECG by the decision rule. Each case was reviewed for acute myocardial infarction as defined by high-grade stenosis on the subsequent emergent cardiac catheterization. A single data collection Microsoft Excel spreadsheet was used, and descriptive statistics were performed in Excel and Stata. Results: Of the 430 CCL activations, 415 (97%; 95% confidence interval, 95%-99%) were identified by the Graff rule. Of the 12 patients who were not identified by the rule, only 2 more were identified by the Glickman criteria. Among patients with confirmed ST-elevation myocardial infarction (79% of CCL activations), the Graff rule was 98% sensitive (95% confidence interval, 96%-99%). Conclusions: The Graff ECG triage rule identified almost all patients for whom the CCL was activated. Modification of the rule as proposed by Glickman added very little to the rule's sensitivity, while increasing the number of ECGs required at triage.

Application-Only Call Graph Construction

Lecture Notes in Computer Science, 2012

Since call graphs are an essential starting point for all interprocedural analyses, many tools an... more Since call graphs are an essential starting point for all interprocedural analyses, many tools and frameworks have been developed to generate the call graph of a given program. The majority of these tools focus on generating the call graph of the whole program (i.e., both the application and the libraries that the application depends on). A popular compromise to the excessive cost of building a call graph for the whole program is to ignore all the effects of the library code and any calls the library makes back into the application. This results in potential unsoundness in the generated call graph and therefore in any analysis that uses it. In this paper, we present Cgc, a tool that generates a sound call graph for the application part of a program without analyzing the code of the library.

Training for Task Specific Keypoint Detection

Lecture Notes in Computer Science, 2009

In this paper, we show that a better performance can be achieved by training a keypoint detector ... more In this paper, we show that a better performance can be achieved by training a keypoint detector to only find those points that are suitable to the needs of the given task. We demonstrate our approach in an urban environment, where the keypoint detector should focus on stable man-made structures and ignore objects that undergo natural changes such as vegetation and clouds. We use Wald-Boost learning with task specific training samples in order to train a keypoint detector with this capability. We show that our aproach generalizes to a broad class of problems where the task is known beforehand.

Joint pose estimator and feature learning for object detection

2009 IEEE 12th International Conference on Computer Vision, 2009

A new learning strategy for object detection is presented. The proposed scheme forgoes the need t... more A new learning strategy for object detection is presented. The proposed scheme forgoes the need to train a collection of detectors dedicated to homogeneous families of poses, and instead learns a single classifier that has the inherent ability to deform based on the signal of interest. Specifically, we train a detector with a standard Ad-aBoost procedure by using combinations of pose-indexed features and pose estimators instead of the usual image features. This allows the learning process to select and combine various estimates of the pose with features able to implicitly compensate for variations in pose. We demonstrate that a detector built in such a manner provides noticeable gains on two hand video sequences and analyze the performance of our detector as these data sets are synthetically enriched in pose while not increased in size.

Toward an Analytical Performance Model to Select between GPU and CPU Execution

2019 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)

Automating the device selection in heterogeneous computing platforms requires the modelling of pe... more Automating the device selection in heterogeneous computing platforms requires the modelling of performance both on CPUs and on accelerators. This work argues for the use of a hybrid analytical performance modelling approach is a practical way to build fast and efficient methods to select an appropriate target for a given computation kernel. The target selection problem has been addressed in the literature, however there has been a strong emphasis on building empirical models with machine learning techniques. We argue that the applicability of such solutions is often limited in production systems. This paper focus on the issue of building a selector to decide if an OpenMP loop nest should be executed in a CPU or in a GPU. To this end, it offers a comprehensive comparison evaluation of the difference in GPU kernel performance on devices of multiple generations of architectures. The idea is to underscore the need for accurate analytical performance models and to provide insights in the evolution of GPU accelerators. This work also highlights a drawback of existing approaches to modelling GPU performance-accurate modelling of memory coalescing characteristics. To that end, we examine a novel application of an inter-thread difference analysis that can further improve analytical models. Finally, this work presents an initial study of an Open Multi-Processing (OpenMP) runtime framework for target-offloading target selection.

Compiler-driven performance

The compiler-driven performance workshop will consist of the presentation of reports on research ... more The compiler-driven performance workshop will consist of the presentation of reports on research progress at various academic and industrial sites across Canada and in the United States. Topics to be discussed in the workshop will include, but are not limited to: • innovative compiler analysis, transformation, and optimization techniques • languages, compilers, and optimization techniques for multicore processors and other parallel architectures • analysis and compilation techniques for dynamic programming languages • compiling for streaming or heterogeneous hardware • dynamic compilation for high-performance and real-time environments • compilation and optimization for scripting languages • compilation techniques for reducing power • program safety • whole system optimization and analysis • tools and infrastructure for compiler research

Supplementary material from "Memory reactivation in rat medial prefrontal cortex occurs in a subtype of cortical UP state during slow-wave sleep

Interaction between hippocampal sharp-wave ripples (SWRs) and UP states, possibly by coordinated ... more Interaction between hippocampal sharp-wave ripples (SWRs) and UP states, possibly by coordinated reactivation of memory traces, is conjectured to play an important role in memory consolidation. Recently, it was reported that SWRs were differentiated into multiple subtypes. However, whether cortical UP states can also be classified into subtypes is not known. Here, we analysed neural ensemble activity from the medial prefrontal cortex from rats trained to run a spatial sequence-memory task. Application of the hidden Markov model (HMM) with three states to epochs of UP-DOWN oscillations identified DOWN states and two subtypes of UP state (UP-1 and UP-2). The two UP subtypes were distinguished by differences in duration, with UP-1 having a longer duration than UP-2, as well as differences in the speed of population vector (PV) decorrelation, with UP-1 decorrelating more slowly than UP-2. Reactivation of recent memory sequences predominantly occurred in UP-2. Short-duration reactivating UP states were dominated by UP-2 whereas long-duration ones exhibit transitions from UP-1 to UP-2. Thus, recent memory reactivation, if it occurred within long-duration UP states, typically was preceded by a period of slow PV evolution not related to recent experience, and which we speculate may be related to previously encoded information. If that is the case, then the transition from UP-1 to UP-2 subtypes may help gradual integration of recent experience with pre-existing cortical memories by interleaving the two in the same UP state.'This article is part of the Theo Murphy meeting issue 'Memory reactivation: replaying events past, present and future'.

CogniCrypt: Supporting developers in using cryptography

2017 32nd IEEE/ACM International Conference on Automated Software Engineering (ASE), 2017

Previous research suggests that developers often struggle using low-level cryptographic APIs and,... more Previous research suggests that developers often struggle using low-level cryptographic APIs and, as a result, produce insecure code. When asked, developers desire, among other things, more tool support to help them use such APIs. In this paper, we present CogniCrypt, a tool that supports developers with the use of cryptographic APIs. CogniCrypt assists the developer in two ways. First, for a number of common cryptographic tasks, CogniCrypt generates code that implements the respective task in a secure manner. Currently, CogniCrypt supports tasks such as data encryption, communication over secure channels, and long-term archiving. Second, CogniCrypt continuously runs static analyses in the background to ensure a secure integration of the generated code into the developer's workspace. This video demo showcases the main features of CogniCrypt: youtube.com/watch?v=JUq5mRHfAWY.

SRAM-PUF-Based Entities Authentication Scheme for Resource-Constrained IoT Devices

IEEE Internet of Things Journal, 2021

With the development of the cloud-based Internet of Things (IoT), people and things can request s... more With the development of the cloud-based Internet of Things (IoT), people and things can request services, access data, or control actuators located thousands of miles away. The entity authentication of the remotely accessed devices is an essential part of the security systems. In this vein, Physical Unclonable Functions (PUFs) are a hot research topic, especially for generating random, stable, and tamper-resistant fingerprints. This paper proposes a lightweight, robust SRAM-PUF based entity authentication scheme to guarantee that the accessed end devices are trustable. The proposed scheme uses Challenge-Response Pairs (CRPs) represented by reordered memory addresses as challenges and the corresponding SRAM cells' startup values as responses. The experimental results show that our scheme can efficiently authenticate resources-constrained IoT devices with a low computation overhead and small memory capacity. Furthermore, we analyze the SRAM-PUF by testing the PUF output under different environmental conditions, including temperature and magnetic field, in addition to exploring the effect of writing different values to the SRAM cells on the stability of their startup values.

Assessment of Sleep, K-Complexes, and Sleep Spindles in a T21 Light-Dark Cycle

Frontiers in Neuroscience, 2020

Circadian rhythm misalignment has a deleterious impact on the brain and the body. In rats, exposu... more Circadian rhythm misalignment has a deleterious impact on the brain and the body. In rats, exposure to a 21-hour day length impairs hippocampal dependent memory. Sleep, and particularly K-complexes and sleep spindles in the cortex, have been hypothesized to be involved in memory consolidation. Altered K-complexes, sleep spindles, or interaction between the cortex and hippocampus could be a mechanism for the memory consolidation failure but has yet to be assessed in any circadian misalignment paradigm. In the current study, continuous local field potential recordings from five rats were used to assess the changes in aspects of behavior and sleep, including wheel running activity, quiet wakefulness, motionless sleep, slow wave sleep, REM sleep, K-complexes and sleep spindles, in rats exposed to six consecutive days of a T21 light-dark cycle (L9:D12). Except for a temporal redistribution of sleep and activity during the T21, there were no changes in period, or total amount for any aspect of sleep or activity. These data suggest that the memory impairment elicited from 6 days of T21 exposure is likely not due to changes in sleep architecture. It remains possible that hippocampal plasticity is affected by experiencing light when subjective circadian phase is calling for dark. However, if there is a reduction in hippocampal plasticity, changes in sleep appear not to be driving this effect.

Lifting the curse of stringly-typed code

How often do JavaScript programmers embed structured languages into strings literals? We conduct ... more How often do JavaScript programmers embed structured languages into strings literals? We conduct an empirical investigating mining nearly 500 thousand JavaScript source files from almost ten thousand repositories from GitHub. We parsed each string literal with seven separate common grammars, and found the most common data type that is hidden within the confines of string literals. To reduce the overuse of strings for structured data types, we present a simple static program analyzer that finds embedded languages and warns the developer, providing an optional fix.

Debugging Static Analysis

IEEE Transactions on Software Engineering, 2019

To detect and x bugs and security vulnerabilities, software companies use static analysis as part... more To detect and x bugs and security vulnerabilities, software companies use static analysis as part of the development process. However, static analysis code itself is also prone to bugs. To ensure a consistent level of precision, as analyzed programs grow more complex, a static analysis has to handle more code constructs, frameworks, and libraries that the programs use. While more complex analyses are written and used in production systems every day, the cost of debugging and xing them also increases tremendously. To better understand the di culties of debugging static analyses, we surveyed 115 static analysis writers. From their responses, we extracted the core requirements to build a debugger for static analysis, which revolve around two main issues: (1) abstracting from two code bases at the same time (the analysis code and the analyzed code) and (2) tracking the analysis internal state throughout both code bases. Most current debugging tools that our survey participants use lack the capabilities to address both issues. Focusing on those requirements, we introduce V , a debugging environment for static data-ow analysis that is integrated in the Eclipse development environment. V features graph visualizations that enable users to view the state of a data-ow analysis and its intermediate results at any time. Special breakpoints in V help users step through the analysis code and the analyzed simultaneously. To evaluate the usefulness of V , we have conducted a user study on 20 static analysis writers. Using V helped our sample of analysis writers identify 25% and x 50% more errors in the analysis code compared to using the standard Eclipse debugging environment. CCS CONCEPTS •Software and its engineering → Software testing and debugging; •Theory of computation → Program analysis; •Humancentered computing → Empirical studies in visualization;

Energy- and Area-Efficient Spin–Orbit Torque Nonvolatile Flip-Flop for Power Gating Architecture

IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2018

This paper proposes two enhanced architectures for spin-orbit torque nonvolatile flip-flop (SOT-N... more This paper proposes two enhanced architectures for spin-orbit torque nonvolatile flip-flop (SOT-NVFF), namely, parallel programming scheme (pSOT-NVFF) and series programming scheme (sSOT-NVFF). At given SOT magnetic tunnel junction (SOT-MTJ) technology, the pSOT-NVFF achieves 29% lower backup energy with nearly 50% shorter backup delay compared with previous parallel architectures and the sSOT-NVFF offers at least 42% lower backup energy with 52% higher speed compared with others. Further analysis of the two proposed architectures showed a preferred window of operation for each architecture as a function of heavy metal electrode resistance (R H M). In particular, sSOT-NVFF is preferred, from speed and energy perspectives, for SOT-MTJs with small R H M (i.e., hundreds of to tens of k), while pSOT-NVFF is preferred for SOT-MTJs with large R H M in the range of hundreds of k. Further study indicates that our NVFF is robust, stable, and functional even with process variations. Index Terms-Nonvolatile flip-flop (NVFF), parallel programming, power gating (PG), series programming, spin-orbit torque magnetic tunnel junction (SOT-MTJ).

Context-, flow-, and field-sensitive data-flow analysis using synchronized Pushdown systems

Proceedings of the ACM on Programming Languages, 2019

Precise static analyses are context-, field- and flow-sensitive. Context- and field-sensitivity a... more Precise static analyses are context-, field- and flow-sensitive. Context- and field-sensitivity are both expressible as context-free language (CFL) reachability problems. Solving both CFL problems along the same data-flow path is undecidable, which is why most flow-sensitive data-flow analyses over-approximate field-sensitivity through k -limited access-path, or through access graphs. Unfortunately, as our experience and this paper show, both representations do not scale very well when used to analyze programs with recursive data structures. Any single CFL-reachability problem is efficiently solvable, by means of a pushdown system. This work thus introduces the concept of synchronized pushdown systems (SPDS). SPDS encode both procedure calls/returns and field stores/loads as separate but “synchronized” CFL reachability problems. An SPDS solves both individual problems precisely, and approximation occurs only in corner cases that are apparently rare in practice: at statements where b...

Towards secure integration of cryptographic software

2015 ACM International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software (Onward!), 2015

While cryptography is now readily available to everyone and can, provably, protect private inform... more While cryptography is now readily available to everyone and can, provably, protect private information from attackers, we still frequently hear about major data leakages, many of which are due to improper use of cryptographic mechanisms. The problem is that many application developers are not cryptographic experts. Even though high-quality cryptographic APIs are widely available, programmers often select the wrong algorithms or misuse APIs due to a lack of understanding. Such issues arise with both simple operations such as encryption as well as with complex secure communication protocols such as SSL. In this paper, we provide a long-term solution that helps application developers integrate cryptographic components correctly and securely by bridging the gap between cryptographers and application developers. Our solution consists of a software product line (with an underlying feature model) that automatically identifies the correct cryptographic algorithms to use, based on the developer's answers to high-level questions in non-expert terminology. Each feature (i.e., cryptographic algorithm) maps into corresponding Java code and a usage protocol describing API restrictions. By composing the user's selected features, we automatically synthesize a secure code blueprint and a usage protocol that corresponds to the selected usage scenario. Since the developer may change the application code over time, we use the usage protocols to statically analyze the program and ensure that the correct use of the API is not violated over time.

Constructing Call Graphs of Scala Programs

Lecture Notes in Computer Science, 2014

Cons ist en t * Comple te * We ll Docum en ted * Easy to * EC OOP * A rt ifact * A EC As Scala ga... more Cons ist en t * Comple te * We ll Docum en ted * Easy to * EC OOP * A rt ifact * A EC As Scala gains popularity, there is growing interest in programming tools for it. Such tools often require call graphs. However, call graph construction algorithms in the literature do not handle Scala features, such as traits and abstract type members. Applying existing call graph construction algorithms to the JVM bytecodes generated by the Scala compiler produces very imprecise results due to type information being lost during compilation. We adapt existing call graph construction algorithms, Name-Based Resolution (RA) and Rapid Type Analysis (RTA), for Scala, and present a formalization based on Featherweight Scala. We evaluate our algorithms on a collection of Scala programs. Our results show that careful handling of complex Scala constructs greatly helps precision and that our most precise analysis generates call graphs with 1.1-3.7 times fewer nodes and 1.5-18.7 times fewer edges than a bytecode-based RTA analysis. 2 Background Algorithms for call graph construction [11] have been studied extensively in the context of object-oriented programming languages such as Java [10, 17], C++ [6] and Self [1], of functional programming languages such as Scheme [24] and ML [12], and of scripting languages such as JavaScript [25]. Roughly speaking, most call graph construction algorithms can be classified as being either type-based or flow-based [7, 13, 14, 17, 18]. The former class of algorithms uses only local information given by static types to determine possible call targets, whereas the latter analyzes the program's data flow. We focus on type-based algorithms, so we will briefly review some important type-based call graph construction algorithms for object-oriented languages upon which our work is based. In the exposition of these algorithms, we use a constraint notation that is equivalent to that of [27], but that explicitly represents call graph edges using a relation ' →' between call sites and methods. Name-Based Resolution (RA). The main challenge in constructing call graphs of object-oriented programs is in approximating the behavior of dynamically dispatched (virtual) method calls. Early work (see, e.g., [26]) simply assumed that a virtual call e.m(• • •) can invoke any method with the same name m. This approach can be captured using the following constraints: main ∈ R RA main c → M M ∈ R RA reachable call c : e.m(.. .) occurs in method M method M has name m M ∈ R c → M RA call Intuitively, rule RA main reads "the main method is reachable" by including it in the set R of reachable methods. Rule RA call states that "if a method is reachable, and a call site c : e.m(.. .) occurs in its body, then every method with name m is reachable from c." Finally, rule RA reachable states that any method M reachable from a call site c is contained in the set R of reachable methods. Class Hierarchy Analysis (CHA). Obviously, Name-Based Resolution can become very imprecise if a class hierarchy contains unrelated methods that happen to have the same name. Class Hierarchy Analysis [9] improves upon name-based resolution by using the static type of the receiver expression of a method call in combination with class hierarchy information to determine what methods may be invoked from a call site. Following the notation of [27], we use StaticType(e) to denote the static type of an expression e, and StaticLookup(C, m) to denote the method definition that is invoked when method m is invoked on an object with run-time type C. Using these definitions, CHA is defined as follows: main ∈ R CHA main c → M M ∈ R CHA reachable call c : e.m(.. .) occurs in method M C ∈ SubTypes(StaticType(e)) StaticLookup(C, m) = M M ∈ R c → M CHA call

An Evaluation of Call Graph Construction Algorithms for Scala

As Scala gains popularity, there is growing interest in programming tools for it. Such tools ofte... more As Scala gains popularity, there is growing interest in programming tools for it. Such tools often require call graphs. However, call graph construction algorithms in the literature do not handle Scala features, such as traits and abstract type members. Applying existing call graph construction algorithms to the JVM bytecodes generated by the Scala compiler produces very imprecise results due to type information being lost during compilation. We adapt existing call graph construction algorithms, Name-Based Resolution (RA) and Rapid Type Analysis (RTA), for Scala, and present a formalization based on Featherweight Scala. We evaluate our algorithms on a collection of Scala programs. Our results show that careful handling of complex Scala constructs greatly helps precision and that our most precise analysis generates call graphs with 1.1-3.7 times fewer nodes and 1.5-18.7 times fewer edges than a bytecode-based RTA analysis.

The Separate Compilation Assumption

Call graphs are an essential requirement for almost all inter-procedural analyses. This motivated... more Call graphs are an essential requirement for almost all inter-procedural analyses. This motivated the development of many tools and frameworks to generate the call graph of a given program. However, the majority of these tools focus on generating the call graph of the whole program (i.e., both the application and the libraries that the application depends on). A popular compromise to the excessive cost of building a call graph for the whole program is to build an application-only call graph. To achieve this, all the effects of the library code and any calls that the library makes back into the application are usually ignored. This results in potential unsoundness in the generated call graph and therefore in analyses that use it. Additionally, the scope of the application classes to be analyzed by such an algorithm has been often arbitrarily defined.

Deconstructing the Deformable Parts Model: Do More with Less

The deformable parts model (DPM) [1] is a successful detection model that continues to achieve st... more The deformable parts model (DPM) [1] is a successful detection model that continues to achieve state-of-the-art results in object detection. While the model's success is attributed to the deformable parts, the proposed system has many other design choices such as the use of multi-resolution HOG features, the sampling of parts from high energy areas, the use of a mixture model, etc. Recently, there have been attempts at analyzing the contribution of these design choices more closely [2]. We delve further into this analysis, performing a study of the effects of restricting the deformation model, the use of single-resolution filters versus multi-resolution filters, and the application of energy "dropout" while sampling parts. Our results indicate that a better performance can be achieved with simpler cost-free deformation models.

FlowBoost — Appearance learning from sparsely annotated video

CVPR 2011, 2011

We propose a new learning method which exploits temporal consistency to successfully learn a comp... more We propose a new learning method which exploits temporal consistency to successfully learn a complex appearance model from a sparsely labeled training video. Our approach consists in iteratively improving an appearancebased model built with a Boosting procedure, and the reconstruction of trajectories corresponding to the motion of multiple targets. We demonstrate the efficiency of our procedure on pedestrian detection in videos and cell detection in microscopy image sequences. In both cases, our method is demonstrated to reduce the labeling requirement by one to two orders of magnitude. We show that in some instances, our method trained with sparse labels on a video sequence is able to outperform a standard learning procedure trained with the fully labeled sequence.

Ability of Triage Decision Rules for Rapid Electrocardiogram to Identify Patients With Suspected ST-Elevation Myocardial Infarction

Critical Pathways in Cardiology: A Journal of Evidence-Based Medicine, 2012

Background: The American College of Cardiology/American Heart Association guidelines for ST-eleva... more Background: The American College of Cardiology/American Heart Association guidelines for ST-elevation myocardial infarction state that an electrocardiogram (ECG) should be performed on patients with suspected acute coronary syndrome upon presentation to the emergency department (ED) within 10 minutes. Objective: To determine how previously published clinical criteria for obtaining an ECG at ED triage perform in a population of patients receiving emergency cardiac catheterization for suspected myocardial infarction. This rule was originally derived by Graff to identify clinical criteria for obtaining an ECG at triage to rapidly identify patients with acute myocardial infarction. The Graff rule was developed in a setting where lytic therapy was the primary reperfusion strategy. A modification proposed by Glickman adds several more criteria in an effort to capture additional patients. We hypothesized that the Graff rule would identify most patients for whom the cardiac catheterization laboratory (CCL) was activated and that the Glickman rule would capture the remaining patients. Methods: Three trained physician reviewers retrospectively applied the Graff decision rule to 430 consecutive patients from a database of emergency CCL activations by ED physicians. The Graff rule recommends that patients between the ages of 30 and 49 years received a rapid ECG if they complained of chest pain and those aged 50 years or older received a rapid ECG when they complained of chest pain, shortness of breath, palpitations, weakness, or syncope. The newly developed Glickman rule, which included nausea and vomiting in patients over the age of 80 years, was applied to the patients where the Graff rule was negative. The triage note or earliest medical contact documentation was used to determine whether the patient's complaints would have resulted in a rapid ECG by the decision rule. Each case was reviewed for acute myocardial infarction as defined by high-grade stenosis on the subsequent emergent cardiac catheterization. A single data collection Microsoft Excel spreadsheet was used, and descriptive statistics were performed in Excel and Stata. Results: Of the 430 CCL activations, 415 (97%; 95% confidence interval, 95%-99%) were identified by the Graff rule. Of the 12 patients who were not identified by the rule, only 2 more were identified by the Glickman criteria. Among patients with confirmed ST-elevation myocardial infarction (79% of CCL activations), the Graff rule was 98% sensitive (95% confidence interval, 96%-99%). Conclusions: The Graff ECG triage rule identified almost all patients for whom the CCL was activated. Modification of the rule as proposed by Glickman added very little to the rule's sensitivity, while increasing the number of ECGs required at triage.

Application-Only Call Graph Construction

Lecture Notes in Computer Science, 2012

Since call graphs are an essential starting point for all interprocedural analyses, many tools an... more Since call graphs are an essential starting point for all interprocedural analyses, many tools and frameworks have been developed to generate the call graph of a given program. The majority of these tools focus on generating the call graph of the whole program (i.e., both the application and the libraries that the application depends on). A popular compromise to the excessive cost of building a call graph for the whole program is to ignore all the effects of the library code and any calls the library makes back into the application. This results in potential unsoundness in the generated call graph and therefore in any analysis that uses it. In this paper, we present Cgc, a tool that generates a sound call graph for the application part of a program without analyzing the code of the library.

Training for Task Specific Keypoint Detection

Lecture Notes in Computer Science, 2009

In this paper, we show that a better performance can be achieved by training a keypoint detector ... more In this paper, we show that a better performance can be achieved by training a keypoint detector to only find those points that are suitable to the needs of the given task. We demonstrate our approach in an urban environment, where the keypoint detector should focus on stable man-made structures and ignore objects that undergo natural changes such as vegetation and clouds. We use Wald-Boost learning with task specific training samples in order to train a keypoint detector with this capability. We show that our aproach generalizes to a broad class of problems where the task is known beforehand.

Joint pose estimator and feature learning for object detection

2009 IEEE 12th International Conference on Computer Vision, 2009

A new learning strategy for object detection is presented. The proposed scheme forgoes the need t... more A new learning strategy for object detection is presented. The proposed scheme forgoes the need to train a collection of detectors dedicated to homogeneous families of poses, and instead learns a single classifier that has the inherent ability to deform based on the signal of interest. Specifically, we train a detector with a standard Ad-aBoost procedure by using combinations of pose-indexed features and pose estimators instead of the usual image features. This allows the learning process to select and combine various estimates of the pose with features able to implicitly compensate for variations in pose. We demonstrate that a detector built in such a manner provides noticeable gains on two hand video sequences and analyze the performance of our detector as these data sets are synthetically enriched in pose while not increased in size.