Resilience

Deep Learning Blindspots

Tools for Fooling the "Black Box"
Saal Adams
Katharine Jarmul
In the past decade, machine learning researchers and theorists have created deep learning architectures which seem to learn complex topics with little intervention. Newer research in adversarial learning questions just how much “learning" these networks are doing. Several theories have arisen regarding neural network “blind spots” which can be exploited to fool the network. For example, by changing a series of pixels which are imperceptible to the human eye, you can render an image recognition model useless. This talk will review the current state of adversarial learning research and showcase some open-source tools to trick the "black box."
This talk aims to: - present recent research on adversarial networks - showcase open-source libraries for fooling a neural network with adversarial learning - recommend possible applications of adversarial networks for social good This talk will include several open-source libraries and research papers on adversarial learning including: Intriguing Properties of neural networks (Szegedy et al., 2013): https://arxiv.org/abs/1312.6199 Explaining and Harnessing Adversarial Examples (Goodfellow et al., 2014) https://arxiv.org/abs/1412.6572 DeepFool: https://github.com/LTS4/DeepFool Deeppwning: https://github.com/cchio/deep-pwning

Additional information

Type lecture
Language English

More sessions

12/27/17
Resilience
Sebastian Jünemann
Saal Borg
Gesundheit als entscheidender Teil von Glück und Zufriedenheit ist bis in ihre kleinsten Teilbereiche „durchkapitalisiert“. Und dieser Prozess macht auch vor humanitärer Hilfe und Krisenintervention nicht halt. In diesem Talk gehen wir auf verschiedene Beispiele ein und erklären, wie CADUS mit seinem Makerspace versucht, dieses Problem auf vielen Ebenen zu hacken.
12/27/17
Resilience
Clifford Wolf
Saal Clarke
Formal hardware verification (hardware model checking) can prove that a design has a specified property. Historically only very simple properties in simple designs have been provable this way, but improvements in model checkers over the last decade enable us to prove very complex design properties nowadays. riscv-formal is a framework for formally verifying RISC-V processors directly against a formal ISA specification. In this presentation I will discuss how the complex task of verifying a ...
12/27/17
Resilience
Alastair Reid
Saal Borg
Formal verification of software has finally started to become viable: we have examples of formally verified microkernels, realistic compilers, hypervisors etc. These are huge achievements and we can expect to see even more impressive results in the future but the correctness proofs depend on a number of assumptions about the Trusted Computing Base that the software depends on. Two key questions to ask are: Are the specifications of the Trusted Computing Base correct? And do the implementations ...
12/27/17
Resilience
Saal Clarke
We shall explain the renewed interest in mix networks. Like Tor, mix networks protect metadata by using layered encryption and routing packets between a series of independent nodes. Mix networks resist vastly more powerful adversary models than Tor though, including global passive adversaries. In so doing, mix networks add both latency and cover traffic. We shall outline the basic components of a mix network, touch on their roles in resisting active and passive attacks, and discuss how the ...
12/28/17
Resilience
raichoo
Saal Dijkstra
Systems are getting increasingly complex and it's getting harder to understand what they are actually doing. Even though they are built by human individuals they often surprise us with seemingly bizarre behavior. DTrace lights a candle in the darkness that is a running production system giving us unprecedented insight into the system helping us to understand what is actually going on. We are going implement `strace`-like functionality, trace every function call in the kernel, watch the scheduler ...
12/28/17
Resilience
Mike Sperber
Saal Dijkstra
Hacker culture overcomes limitations in computer systems through creativity and tinkering. At the same time, hacker culture has shaped the practice of software development to this day. This is problematic - techniques effective for breaking (into) a computer systems are not necessarily suitable for developing resilient and secure systems. It does not have to be this way: We can approach software development as a methodical, systematic activity rather than tinkering, and teach it accordingly. ...
12/29/17
Resilience
Trammell Hudson
Saal Clarke
The NERF and Heads projects bring Linux back to the cloud servers' boot ROMs by replacing nearly all of the vendor firmware with a reproducible built Linux runtime that acts as a fast, flexible, and measured boot loader. It has been years since any modern servers have supported Free Firmware options like LinuxBIOS or coreboot, and as a result server and cloud security has been dependent on unreviewable, closed source, proprietary vendor firmware of questionable quality. With Heads on NERF, we ...