Exploring the boundary between accuracy and performances in recurrent neural networks

Exploring the boundary between accuracy and performances in recurrent neural networks

When it comes to interpreting streams of data using modern artificial intelligence techniques, such as audio in speech recognition, computational requirements of state-of- the-art models can easily skyrocket and result in huge power requirements. However, accepting a small loss in their accuracy can go a long way in reducing their resource impact. This work explores the boundary between accuracy and performances in such a context. By Alessandro Pappalardo PhD student @Politecnico di Milano   Modern artificial intelligence approaches to problems such as image captioning, i.e. describing content of an image, can already have significant computational requirements. On top of that, scaling such techniques from processing a single data point, e.g. an image, to a sequence of them, e.g. a video, increases their requirements non-linearly. The reason is that interpreting a…
Read More
HyPPO: Hybrid Performance-aware Power-capping Orchestration in containerized environments

HyPPO: Hybrid Performance-aware Power-capping Orchestration in containerized environments

Energy proportionality represents the key aspect in order to reduce the TCO (Total Cost of Ownership) in modern datacenters and on premise systems. HyPPO achieves energy proportionality allowing energy awarness and autonomic management of containerized environment based on Kubernetes. By Marco Arnaboldi PhD student @Politecnico di Milano   In the last decade cloud computing became the go-to choice for companies and developers to deploy, manage and maintain web services at scale. In this context, Docker containers are now becoming the de-facto standard for cloud native application due to their flexibility and their ability to introduce faster development and deployment cycles. Among the various workloads running inside cloud platforms, an interesting category is represented by On-Line Data Intensive (OLDI) workloads. This kind of workloads are typically composed of large deployments with…
Read More
Exploiting FPGA from Data Science Programming Languages

Exploiting FPGA from Data Science Programming Languages

This work presents different methodologies to create Hardware Libraries for FPGAs. This allows data scientists and software developers to use such devices transparently from Matlab, Python and R applications, running on Desktop or Embedded systems. By Luca Stornaiuolo PhD student @Politecnico di Milano   In the last years, the huge amount of available data leads data scientists to look for increasingly powerful systems to process them. Within this context, Field Programmable Gate Arrays (FPGAs) are a promising solution to improve the performance of the system while keeping low the energy consumption. Nevertheless, exploiting FPGAs is very challenging due to the high level of expertise required to program them. A lot of High Level Synthesis tools have been produced to help programmers during the flow of acceleration of their algorithms through…
Read More
DEEP-mon: monitoring your data-center power consumption

DEEP-mon: monitoring your data-center power consumption

Docker containers are spreading and becoming the reference tool for managing applications in the cloud. DEEP-mon efficiently monitors performance and power of containers to enable energy awareness and autonomic management of the next generation of cloud computing systems. By Rolando Brondolin PhD student @Politecnico di Milano   Data-centers and cloud computing are in our everyday life even if we don’t always see them. From the instant message we sent this morning with our smartphone to car sharing, e-commerce and social media, most of our interactions with the web are powered by cloud computing. In this context, performance and energy efficiency are two important aspects of this web revolution. On the one hand, performance means speed and ease of use for the applications that we use in our everyday life. On the…
Read More
Enhancing Personalized Medicine Research with a HUG

Enhancing Personalized Medicine Research with a HUG

Hardware for hUman Genomics (HUG) is a framework that exploits hardware accelerators (FPGAs) to speedup research in the field of personalized medicine. Its high level of abstraction allows researchers and doctors to exploit the potentiality of hardware accelerators to create drugs shaped on the DNA of the individual. By Lorenzo Di Tucci PhD student @Politecnico di Milano In the coming years, human genome research will likely transform medical practices. The unique genetic profile of an individual and the knowledge of molecular basis of diseases are leading to the development of personalized medicines and therapies, but the exponential growth of available genomic data requires a computational effort that may limit the progress of personalized medicine. Within this context, HUG is a novel hardware and software integrated system developed at NECSTLab (Politecnico…
Read More
Arancino: A tasty framework to measure Evasive Malware

Arancino: A tasty framework to measure Evasive Malware

Malware authors have been developing techniques with the purpose of hiding their creations from Analysis  platforms. Hence, we performed a measurement of which of those are used to detect and break analysis systems based on Dynamic Binary Instrumentation (DBI) Tools. By Mario Polino Postdoc researcher @Politecnico di Milano After decades of research and development, the problem of Malware still persists. Certainly, the approach and the motivation behind malware creation and spread are changed. Nowadays there are several platforms and advanced systems that can perform accurate analysis on software and, in particular, on malware samples. For this reason, malware authors have been developing techniques to hide their creations from such platforms. E.g., They conduct some tests to check if the malware sample is running under a virtual machine or if the list,…
Read More
CONDOR: Convolutional Neural Networks Dataflow Optimization Using Reconfigurable Hardware

CONDOR: Convolutional Neural Networks Dataflow Optimization Using Reconfigurable Hardware

Condor is an end-to-end framework to implement Convolutional Neural Networks on FPGA, that does not require the user to have experience in FPGA programming. The framework is able to interpret models from the well-known deep learning engine Caffe. By Giuseppe Natale PhD student @Politecnico di Milano The recent years have seen a rapid diffusion of deep learning algorithms as Convolutional Neural Networks (CNNs) and, as a consequence, an intensification of industrial and academic research focused on optimizing their implementation. Different computing architectures have been explored and, among all of them, Field Programmable Gate Arrays (FPGAs) seem to be a very attractive choice, since they can deliver sustained performances with high power efficiency, as CNNs can be directly mapped onto hardware and still offer flexibility thanks to their programmability. Nevertheless, the…
Read More
I’m not malicious, detection of evasive Android malware

I’m not malicious, detection of evasive Android malware

The increasing popularity of the smartphones attracted lots “bad actors” that wants to spread malicious software into the ecosystem for profit. To avoid being detected and maximize profit, malware uses evasive techniques. We propose an approach to combat evasive malware. By Chengyu Zheng PhD student @Politecnico di Milano How to avoid being detected With over 500 million devices and an estimated 84% market share, Android-based devices are the main target for cyber-criminals. In addition to the alarming amount of malware families and samples, evasive techniques used by malwares are becoming more and more sophisticated. With the high amount of new applications being released every month, “app store” maintainer are struggling to find a reliable solutions to analyze apps in order to recognize and isolate malicious ones. Techniques used to analyze…
Read More
FROST: a common backend to accelerate Domain Specific Languages on FPGA

FROST: a common backend to accelerate Domain Specific Languages on FPGA

Domain Specific Languages are gaining more and more interest thanks to the significant level of performance they can reach on different architectures. FROST is a common backend able to accelerate on FPGA applications developed in different DSLs.   By Emanuele Del Sozzo Ph.D. student @ Politecnico di Milano Due to the reaching of the end of Dennard scaling and Moore’s law, we are experiencing a growing interest towards Heterogeneous System Architectures (HSAs) as a promising solution to boost performance and, at the same time, reduce power consumption. The combination of different hardware accelerators, like GPUs, FPGAs, and ASICs, along with CPUs, allows to choose the most suitable architecture for a specific task, and, for this reason, many high-performance systems are currently taking advantage of heterogeneity. [caption id="attachment_506" align="aligncenter" width="300"] Example…
Read More
Breaking… the laws of robotics: attacking industrial robots

Breaking… the laws of robotics: attacking industrial robots

Industrial robots are everywhere: what happens if they get compromised? Is this hard? Are they attractive for attackers? How can we improve their security? To answer these questions, last year we studied the security landscape of an industrial robot and we analysed (and compromised) a widespread robot.   By Marcello Pogliani PhD student at the NECSTLab, working on Systems Security Industrial robots are drastically evolving: on one side, “caged” giant robots are being complemented by smaller, “collaborative” models designed to share the workspace with human workers; on the other side, they are more “intelligent”, for example, by means of an improved interconnection for tasks such as remote maintenance, and integration with information systems. This means that robots, once “air-gapped”, are now exposed to hostile avenues. What happens (Skynet aside) if…
Read More