Deep learning is a form of machine learning that enables computers to learn from experience and understand the world in terms of a hierarchy of concepts. Because the computer gathers knowledge from experience, there is no need for a human computer operator to formally specify all the knowledge that the computer needs. The hierarchy of concepts allows the computer to learn complicated concepts by building them out of simpler ones; a graph of these hierarchies would be many layers deep. This book introduces a broad range of topics in deep learning. The text offers mathematical and conceptual background, covering relevant concepts in linear algebra, probability theory and information theory, numerical computation, and machine learning. It describes deep learning techniques used by practitioners in industry, including deep feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling, and practical methodology; and it surveys such applications as natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and videogames. Finally, the book offers research perspectives, covering such theoretical topics as linear factor models, autoencoders, representation learning, structured probabilistic models, Monte Carlo methods, the partition function, approximate inference, and deep generative models. Deep Learning can be used by undergraduate or graduate students planning careers in either industry or research, and by software engineers who want to begin using deep learning in their products or platforms. A website offers supplementary material for both readers and instructors.

Deep Learning

On robust estimation of the location parameter

Artificial Neural Networks have dramatically improved performance for many machine learning tasks. We demonstrate a new architecture for a fully optical neural network that enables a computational speed enhancement of at least two orders of magnitude and three orders of magnitude in power efficiency over state-of-the-art electronics.

Deep learning with coherent nanophotonic circuits

Most of the signal processing that we will study in this course involves local operations on a signal, namely transforming the signal by applying linear combinations of values in the neighborhood of each sample point. You are familiar with such operations from Calculus, namely, taking derivatives and you are also familiar with this from optics namely blurring a signal. We will be looking at sampled signals only. Let's start with a few basic examples. Local difference Suppose we have a 1D image and we take the local difference of intensities, DI(x) = 1 2 (I(x + 1) − I(x − 1)) which give a discrete approximation to a partial derivative. (We compute this for each x in the image.) What is the effect of such a transformation? One key idea is that such a derivative would be useful for marking positions where the intensity changes. Such a change is called an edge. It is important to detect edges in images because they often mark locations at which object properties change. These can include changes in illumination along a surface due to a shadow boundary, or a material (pigment) change, or a change in depth as when one object ends and another begins. The computational problem of finding intensity edges in images is called edge detection. We could look for positions at which DI(x) has a large negative or positive value. Large positive values indicate an edge that goes from low to high intensity, and large negative values indicate an edge that goes from high to low intensity. Example Suppose the image consists of a single (slightly sloped) edge:

http://ieeexplore.ieee.org/iel5/5/31307/01456462.pdf

Linear systems

Recent advances in physical reservoir computing: A review

https://www.cse.iitk.ac.in/users/isaha/Publications/Journals/NC10.pdf/

Artificial neural networks in hardware: A survey of two decades of progress

Author(s): Saha, I; Ramaithitima, R; Kumar, V; Pappas, GJ; Seshia, SA | Abstract: We present a compositional motion planning framework for multi-robot systems based on an encoding to satisfiability modulo theories (SMT). In our framework, the desired behavior of a group of robots is specified using a set of safe linear temporal logic (LTL) properties. Our method relies on a library of motion primitives, each of which corresponds to a controller that ensures a particular trajectory in a given configuration. Using the closed-loop behavior of the robots under the action of different controllers, we formulate the motion planning problem as an SMT solving problem and use an off-the-shelf SMT solver to generate trajectories for the robots. Our approach can also be extended to synthesize optimal cost trajectories where optimality is defined with respect to the available motion primitives. Experimental results show that our framework can efficiently solve complex motion planning problems in the context of multi-robot systems.

/pdf/automated-composition-of-motion-primitives-for-multi-robot-1218qnij61.pdf

Automated composition of motion primitives for multi-robot systems from safe LTL specifications

Distributed mobile robotics (DMR) involves teams of networked robots navigating in a physical space to achieve tasks in a coordinated fashion. A major challenge in DMR is to program the ensemble of robots with formal guarantees and high assurance of correct operation. To this end, we introduce DRONA, a framework for building reliable DMR applications. This paper makes three central contributions: (1) We present a novel and provably correct decentralized asynchronous motion planner that can perform on-the-fly collision-free planning for dynamically generated tasks. Moreover, the motion planner is the first to take into account the fact that distributed robots may have clocks that are only synchronized up to a tolerance, i.e., they are almost synchronous; (2) We formalize the DMR system as a mixed-synchronous system, and present a sound abstraction-based verification approach for DMR systems, and (3) DRONA provides a state-machine based language for safe event-driven programming of a DMR system and the code generated by the compiler can be executed on platforms such as the robot operating system (ROS). To demonstrate the efficacy of DRONA, we build and verify a priority mail delivery system. Using our abstraction-based verification approach we were able to find, within a few minutes, bugs which could not be found by performing random simulation for several hours. Our verified decentralized motion-planner scales efficiently for large number of robots (upto 128 robots) and workspace sizes (upto a 256x256 grid).

/pdf/drona-a-framework-for-safe-distributed-mobile-robotics-1xf8l5z3sy.pdf

DRONA: a framework for safe distributed mobile robotics

Software implementations of controllers for physical subsystems form the core of many modern safety-critical systems such as aircraft flight control and automotive engine control. A fundamental property of such implementations is stability, the guarantee that the physical plant converges to a desired behavior under the actions of the controller. We present a methodology and a tool to perform automated static analysis of embedded controller code for stability of the controlled physical system.The design of controllers for physical systems provides not only the controllers but also mathematical proofs of their stability under idealized mathematical models. Unfortunately, since these models do not capture most of the implementation details, it is not always clear if the stability properties are retained by the software implementation, either because of software bugs, or because of imprecisions arising from fixed-precision arithmetic or timing.Our methodology is based on the following separation of concerns. First, we analyze the controller mathematical models to derive bounds on the implementation errors that can be tolerated while still guaranteeing stability. Second, we automatically analyze the controller software to check if the maximal implementation error is within the tolerance bound computed in the first step.We have implemented this methodology in Costan, a tool to check stability for controller implementations. Using Costan, we analyzed a set of control examples whose mathematical models are given in Matlab/Simulink and whose C implementation is generated using Real-Time Workshop. Unlike previous static analysis research, which has focused on proving low-level runtime properties such as absence of buffer overruns or arithmetic overflows, our technique combines analysis of the mathematical controller models and automated analysis of source code to guarantee application-level stability properties.

Automatic verification of control system implementations

A key feature of control systems is robustness, the property that small perturbations in the system inputs cause only small changes in its outputs. Robustness is key to designing systems that work under uncertain or imprecise environments. While continuous control design algorithms can explicitly incorporate robustness as a design goal, it is not clear if robustness is maintained at the software implementation level of the controller: two ``close'' inputs can execute very different code paths which may potentially produce vastly different outputs. We present an algorithm and a tool to characterize the robustness of a control software implementation. Our algorithm is based on symbolic execution and non-linear optimization, and computes the maximum difference in program outputs over all program paths when a program input is perturbed. As a by-product, our algorithm generates a set of test vectors which demonstrate the worst-case deviations in outputs for small deviations in inputs. We have implemented our approach on top of the Splat test generation tool and we describe an evaluation of our implementation on two examples of automotive control code.

Indranil Saha

Papers

Artificial neural networks in hardware: A survey of two decades of progress

Automated composition of motion primitives for multi-robot systems from safe LTL specifications

DRONA: a framework for safe distributed mobile robotics

Automatic verification of control system implementations

Symbolic Robustness Analysis