Personal tools

Difference between revisions of "IBM Research"

From iis-projects

Jump to: navigation, search
(Useful Reading)
(Contact)
 
(50 intermediate revisions by 5 users not shown)
Line 3: Line 3:
 
[[Category:Master Thesis]]  
 
[[Category:Master Thesis]]  
 
[[Category:Available]]  
 
[[Category:Available]]  
[[Category:2020]]
+
[[Category:2021]]
 
[[Category:Hot]]
 
[[Category:Hot]]
 
[[File:IBM_ZRLab.png|center]]
 
[[File:IBM_ZRLab.png|center]]
Line 20: Line 20:
 
The location in Zurich is one of IBM’s 12 global research labs. IBM has maintained a research laboratory in Switzerland since 1956. As the first European branch of IBM Research, the mission of the Zurich Lab, in addition to pursuing cutting-edge research for tomorrow’s information technology, is to cultivate close relationships with academic and industrial partners, be one of the premier places to work for world-class researchers, to promote women in IT and science, and to help drive Europe’s innovation agenda. [https://www.zurich.ibm.com/pdf/ZRL_Fact_Sheet.pdf Download factsheet]
 
The location in Zurich is one of IBM’s 12 global research labs. IBM has maintained a research laboratory in Switzerland since 1956. As the first European branch of IBM Research, the mission of the Zurich Lab, in addition to pursuing cutting-edge research for tomorrow’s information technology, is to cultivate close relationships with academic and industrial partners, be one of the premier places to work for world-class researchers, to promote women in IT and science, and to help drive Europe’s innovation agenda. [https://www.zurich.ibm.com/pdf/ZRL_Fact_Sheet.pdf Download factsheet]
  
==Hyperdimensional Computing (HDC)==
+
==Hybrid AI Systems (HAS)==
  
 
[[File:NatureElectronics20.jpg|thumb|right|200px]]
 
[[File:NatureElectronics20.jpg|thumb|right|200px]]
Hyperdimensional computing (HDC) is a brain-inspired computing paradigm based on representing information with hypervectors with dimensions in the thousands. Hypervectors are holographic and (pseudo)random with independent and identically distributed (i.i.d.) components. Principles of HDC allow implementing efficient machine learning models, where it has shown few-shot learning capabilities by requiring less training data than conventional machine learning models to do accurate predictions. By its very nature, HDC is extremely robust against failures, defects, variations, and noise, all of which are complementary to ultra-low energy computation on nanoscale fabrics such as Phase-Change Memory (PCM) devices.  By eliminating non von-Neumann bottleneck, PCM-based in-memory computing systems enable building systems that can run HDC models at very high energy efficiencies. This opens doors to a vast spectrum of further research in applications, algorithms, architecture and system development.
+
Neither symbolic AI nor neural networks alone has produced the kind of intelligence expressed in human and animal behavior. Why? Each has a long and rich history, but has addressed a relatively narrow aspect of the problem. Symbolic AI focuses on solving cognitive problems, drawing upon the rich framework of symbolic computation to manipulate internal representations in order to perform reasoning and inference. But it suffers from being non-adaptive, lacking the ability to learn from example or by direct observation of the world (aka symbol grounding problem). Neural networks on the other hand have the ability to learn from data, and derive much of their power from nonlinear function approximation combined with stochastic gradient descent. But intelligence requires more than modeling input-output relationships. Without the richness of symbolic computation, neural nets lack the simple but powerful operations such as variable binding that allow for analogy making and reasoning, which underlie the ability to generalize from few examples. To address this gap, '''neurosymbolic AI''' aims to combine the best of both worlds to approach human-level intelligence.
  
 
+
We approach the problem from a very different perspective, inspired by the brain’s high-dimensional circuits and the unique mathematical properties of high-dimensional spaces. ''It leads us to a novel information processing architecture that combines the strengths of symbolic AI and neural networks, and yet has novel emergent properties of its own.'' By combining a small set of basic operations on high-dimensional vectors, we obtain hybrid AI system (HAS) that makes it possible to represent and manipulate data in ways familiar to us from symbolic AI, and to learn from the statistics of data in ways familiar to us from artificial neural networks and deep learning. Further, principles of such HAS allow few-shot learning capabilities, and extremely robust operations against failures, defects, variations, and noise, all of which are complementary to ultra-low energy computation on nanoscale fabrics such as phase-change memory devices. Exciting further research (listed in below table) awaiting in this direction spans high-level algorithmic exploration all the way to efficient hardware design for emerging computational fabrics.
HDC has also been successfully applied in few-shot meta-learning problems - where a controller is trained to distinguish previously unseen classes by feeding only a few examples. The state-of-the-art accuracy in few-shot learning problems has been achieved by augmenting the controller with an explicit memory and guiding the controller to learning HDC representations. Exciting further research awaiting in this direction includes exploring compact representation of explicit memory and exploring novel cost functions and training algorithms to further improving the classification accuracy.
 
  
 
===Useful Reading===
 
===Useful Reading===
*[https://link.springer.com/article/10.1007/s12559-009-9009-8 Hyperdimensional Computing: An Introduction to Computing in Distributed Representation with High-Dimensional Random Vectors]
+
*[https://www.youtube.com/watch?v=HhymId8dr5Q Neurosymbolic AI Explained, IBM-Research]
*[https://www.nature.com/articles/s41928-020-0410-3 In-memory hyperdimensional computing]
+
*[https://www.youtube.com/watch?v=CYbkoAP5dME Neurosymbolic AI, Invited Talk by David Cox, IAAI /AAAI 2020]  
*[https://arxiv.org/pdf/2010.01939.pdf Robust High-dimensional Memory-augmented Neural Networks]
+
*[https://www.nature.com/articles/s41928-020-0410-3 In-memory hyperdimensional computing], Nature Electronics, 2020.
 +
*[https://www.nature.com/articles/s41467-021-22364-0 Robust high-dimensional memory-augmented neural networks], Nature Communications, 2021.
 +
*[https://www.nature.com/articles/s41565-023-01357-8 In-memory factorization of holographic perceptual representations], Nature Nanotechnology, 2023.
  
 
===Prerequisites===
 
===Prerequisites===
Line 37: Line 38:
 
*Background in machine learning (''recommended'')
 
*Background in machine learning (''recommended'')
 
*Experience with any deep learning framework such as TensorFlow or PyTorch (''recommended'')
 
*Experience with any deep learning framework such as TensorFlow or PyTorch (''recommended'')
*VLSI I and VLSI II (''recommended'')
+
*VLSI I (''recommended'')
 +
 
 +
<!-- ==In-Memory Computing (IMC)==
  
==In-Memory Computing (IMC)==
+
[[File:NNcover_imc.jpg|thumb|right|200px]]
 
For decades, conventional computers based on the von Neumann architecture have performed computation by repeatedly transferring data between their processing and their memory units, which are physically separated. As computation becomes increasingly data-centric and as the scalability limits in terms of performance and power are being reached, alternative computing paradigms are searched for in which computation and storage are collocated. A fascinating new approach is that of computational memory where the physics of nanoscale memory devices are used to perform certain computational tasks within the memory unit in a non-von Neumann manner. '''Computational Memory''' (CM) is finding application in a variety of areas such as machine learning and signal processing. In addition to novel non-volatile memory technologies like '''PCM''' and '''ReRAM''', also conventional '''SRAM''' has been proposed as underlying storage technology in Computational Memories.
 
For decades, conventional computers based on the von Neumann architecture have performed computation by repeatedly transferring data between their processing and their memory units, which are physically separated. As computation becomes increasingly data-centric and as the scalability limits in terms of performance and power are being reached, alternative computing paradigms are searched for in which computation and storage are collocated. A fascinating new approach is that of computational memory where the physics of nanoscale memory devices are used to perform certain computational tasks within the memory unit in a non-von Neumann manner. '''Computational Memory''' (CM) is finding application in a variety of areas such as machine learning and signal processing. In addition to novel non-volatile memory technologies like '''PCM''' and '''ReRAM''', also conventional '''SRAM''' has been proposed as underlying storage technology in Computational Memories.
  
 
===Useful Reading===
 
===Useful Reading===
 +
*[https://ieeexplore.ieee.org/document/9288639 An SRAM-Based Multibit In-Memory Matrix-Vector Multiplier With a Precision That Scales Linearly in Area, Time, and Power]
 
*[https://www.nature.com/articles/s41565-020-0655-z Memory devices and applications for in-memory computing]
 
*[https://www.nature.com/articles/s41565-020-0655-z Memory devices and applications for in-memory computing]
 
*[https://ieeexplore.ieee.org/abstract/document/8865099 Deep learning acceleration based on in-memory computing]
 
*[https://ieeexplore.ieee.org/abstract/document/8865099 Deep learning acceleration based on in-memory computing]
Line 49: Line 53:
 
*General interest in Deep Learning and memory/system design
 
*General interest in Deep Learning and memory/system design
 
*VLSI I and VLSI II (''recommended'')
 
*VLSI I and VLSI II (''recommended'')
*Analog Circuit Design (''recommended'')
 
 
Specific requirements for the different projects vary and are generally negotiable.
 
Specific requirements for the different projects vary and are generally negotiable.
 +
--->
  
 
==Available Projects==
 
==Available Projects==
We are inviting applications from students to conduct their master’s thesis work or an internship project at the IBM Research lab in Zurich on this exciting new topic.  
+
We are inviting applications from students to conduct their thesis (bachelor, semester, and master) or an internship project at the IBM Research lab in Zurich on this exciting new topic.  
 
<!--
 
<!--
 
The work performed could span low-level hardware experiments on phase-change memory chips comprising more than 1 million devices to high-level algorithmic development in a deep learning framework such as TensorFlow or PyTorch. It also involves interactions with several researchers across IBM research focusing on various aspects of the project. The ideal candidate should have a multi-disciplinary background, strong mathematical aptitude and programming skills. Prior knowledge on emerging memory technologies such as phase-change memory is a bonus but not necessary.
 
The work performed could span low-level hardware experiments on phase-change memory chips comprising more than 1 million devices to high-level algorithmic development in a deep learning framework such as TensorFlow or PyTorch. It also involves interactions with several researchers across IBM research focusing on various aspects of the project. The ideal candidate should have a multi-disciplinary background, strong mathematical aptitude and programming skills. Prior knowledge on emerging memory technologies such as phase-change memory is a bonus but not necessary.
Line 60: Line 64:
 
{| class="wikitable" style="text-align: center;"
 
{| class="wikitable" style="text-align: center;"
 
|-
 
|-
! style="width: 5%;"|Type !! style="width: 20%"|Project !! style="width: 60%"|Description !! style="width: 5%"|Topic !! style="width: 15%"|Workload Type  
+
! style="width: 5%;"|Type !! style="width: 20%"|Project !! style="width: 60%"|Description !! style="width: 5%"|Workload Type  
 
|-
 
|-
| MA|| Lifelong learning challenge || [http://iis-projects.ee.ethz.ch/images/0/01/IBM_MANN_Y2020.pdf '''Artificial general intelligence (AGI):''' lifelong learning challenge] || HDC || algorithmic design
+
 
 +
 
 +
 
 +
| MA|| Code Generation for Analog In- Memory Computing SoCs || [https://iis-projects.ee.ethz.ch/images/4/48/Y2024_IBM_CodeGen_MasterThesis.pdf Link to description] || algorithmic design  
 
|-
 
|-
| MA|| Machine learning based on optimal transport using in-memory computing' ||  [http://iis-projects.ee.ethz.ch/images/3/33/IBM_OT_Y2020.pdf '''Machine learning based on optimal transport using in-memory computing'''] || HDC || algorithmic design
+
 
 +
 
 +
 
 +
| MA || Neurosymbolic Architectures to Approach Human-like AI ||  [https://iis-projects.ee.ethz.ch/images/c/cc/IBM_Neurosymbolic.pdf Link to description] || algorithmic design
 
|-
 
|-
| MA || Developing Efficient Models of Strong AI for Intelligence Quotient (IQ) Test ||  [http://iis-projects.ee.ethz.ch/images/4/4b/IBM_RPM.pdf Developing Efficient Models of '''Strong AI for Intelligence Quotient (IQ) Test'''|| HDC || algorithmic design
+
 
 +
 
 +
 
 +
| MA || Developing Efficient Models for Solving Intelligence Quotient (IQ) Test ||  [https://iis-projects.ee.ethz.ch/images/4/4b/IBM_RPM.pdf Link to description]  || algorithmic/hardware design
 
|-
 
|-
| MA || Accurate deep learning inference using computational memory || [https://www.zurich.ibm.com/careers/2020_027.html Project description and application] || IMC || algorithmic design
+
 
 +
 
 +
| MA|| Crytography meets in-memory computing || [https://iis-projects.ee.ethz.ch/images/6/62/Y2021_10_Advertisement_VME.pdf Link to description] || algorithmic design and hardware experiments
 
|-
 
|-
| MA || ADC design for computational memory || style="text-align: justify;"|Digital-to-Analog converters ('''DAC'''s) and Analog-to-Digital converters ('''ADC'''s) are extensively employed in Computational Memory ('''CM''')  to handle the crossing between the digital and analog domain, in which computationally expensive tasks, like Matrix-Vector Multiplications ('''MVM'''), are carried out with O(1) complexity. Each conversion costs a certain amount of energy and its precision can only be guaranteed up to the Effective Number of Bits ('''ENOB''') of the employed data converter. <br /> The research focus will be on understanding the system level requirements on ADC and DAC for optimal performance of Deep Neural Network inference using CM. Furthermore, the effects of noise, non-linearity and manufacturing tolerances shall be examined and counter measurements, like for example periodic digital ADC recalibration and digital post processing, shall be evaluated with regards to effectivity and energy costs.|| IMC || analog circuit design
+
 
 +
 
 +
| MA|| Optimal routing for 2D Mesh-based Analog Compute-In-Memory Accelerator Architecture || [https://iis-projects.ee.ethz.ch/images/8/83/ADS_Routing.pdf Link to description] || hardware design  
 
|-
 
|-
| SA || Testing of a computational memory chip || style="text-align: justify;"| This project is about building a Microprocessor/FPGA-based test platform around a novel IMC chip. After commissioning the chip, DL workload tests can be run to characterize its throughput and energy-efficiency.|| IMC || PCB Design <br /> and/or<br />  µ-code implementation
+
 
|-
+
 
| MA/SA || Neural Network Training on a <br /> computational memory chip || style="text-align: justify;"|TBA || IMC || algorithm/system design<br />and/or<br />  analog circuit design
+
 
 
|}
 
|}
  
 
==Contact==
 
==Contact==
 
: Thesis will be at IBM Zurich in Rüschlikon
 
: Thesis will be at IBM Zurich in Rüschlikon
; Hyperdimensional Computing (HDC) projects
+
; Hybrid AI Systems (HAS) projects
: Contact (at ETH Zurich): [[:User:kgf | Dr. Frank K. Gurkaynak]] and [[:User:Herschmi | Michael Hersche]]
+
: Contact (at ETH Zurich): [[:User:kgf | Dr. Frank K. Gurkaynak]]
 +
: Contact (at IBM): [mailto:ibo@zurich.ibm.com Dr. Irem Boybat]
 +
: Contact (at IBM): [mailto:anu@zurich.ibm.com Dr. Manuel Le Gallo]  
 
: Contact (at IBM): [mailto:ase@zurich.ibm.com Dr. Abu Sebastian]  
 
: Contact (at IBM): [mailto:ase@zurich.ibm.com Dr. Abu Sebastian]  
 
: Contact (at IBM): [mailto:abr@zurich.ibm.com Dr. Abbas Rahimi]
 
: Contact (at IBM): [mailto:abr@zurich.ibm.com Dr. Abbas Rahimi]
: Professor: [http://www.iis.ee.ethz.ch/portrait/staff/lbenini.en.html Luca Benini]
+
: Professor: [https://iis.ee.ethz.ch/people/person-detail.html?persid=194234 Luca Benini]
; In-Memory Computing (IMC) projects
 
: Contact (at IBM/ETH Zurich): [[:User:rid | Riduan Khaddam-Aljameh]]
 
: Contact (at IBM): [mailto:ase@zurich.ibm.com Dr. Abu Sebastian]
 
: Professor: [http://www.iis.ee.ethz.ch/portrait/staff/lbenini.en.html Luca Benini]
 

Latest revision as of 14:20, 15 March 2024

IBM ZRLab.png

Short Description

Today, we are entering the era of cognitive computing, which holds great promise in deriving intelligence and knowledge from huge volumes of data. In today’s computers based on von Neumann architecture, huge amounts of data need to be shuttled back and forth at high speeds, a task at which this architecture is inefficient.

It is becoming increasingly clear that to build efficient cognitive computers, we need to transition to non-von Neumann architectures in which memory and processing coexist in some form. At IBM Research–Zurich in the Neuromorphic and In-memory Computing Group, we explore various such computing paradigms from in-memory computing to brain-inspired neuromorphic computing. Our research spans from devices and architectures to algorithms and applications.

About the IBM Research–Zurich

The location in Zurich is one of IBM’s 12 global research labs. IBM has maintained a research laboratory in Switzerland since 1956. As the first European branch of IBM Research, the mission of the Zurich Lab, in addition to pursuing cutting-edge research for tomorrow’s information technology, is to cultivate close relationships with academic and industrial partners, be one of the premier places to work for world-class researchers, to promote women in IT and science, and to help drive Europe’s innovation agenda. Download factsheet

Hybrid AI Systems (HAS)

NatureElectronics20.jpg

Neither symbolic AI nor neural networks alone has produced the kind of intelligence expressed in human and animal behavior. Why? Each has a long and rich history, but has addressed a relatively narrow aspect of the problem. Symbolic AI focuses on solving cognitive problems, drawing upon the rich framework of symbolic computation to manipulate internal representations in order to perform reasoning and inference. But it suffers from being non-adaptive, lacking the ability to learn from example or by direct observation of the world (aka symbol grounding problem). Neural networks on the other hand have the ability to learn from data, and derive much of their power from nonlinear function approximation combined with stochastic gradient descent. But intelligence requires more than modeling input-output relationships. Without the richness of symbolic computation, neural nets lack the simple but powerful operations such as variable binding that allow for analogy making and reasoning, which underlie the ability to generalize from few examples. To address this gap, neurosymbolic AI aims to combine the best of both worlds to approach human-level intelligence.

We approach the problem from a very different perspective, inspired by the brain’s high-dimensional circuits and the unique mathematical properties of high-dimensional spaces. It leads us to a novel information processing architecture that combines the strengths of symbolic AI and neural networks, and yet has novel emergent properties of its own. By combining a small set of basic operations on high-dimensional vectors, we obtain hybrid AI system (HAS) that makes it possible to represent and manipulate data in ways familiar to us from symbolic AI, and to learn from the statistics of data in ways familiar to us from artificial neural networks and deep learning. Further, principles of such HAS allow few-shot learning capabilities, and extremely robust operations against failures, defects, variations, and noise, all of which are complementary to ultra-low energy computation on nanoscale fabrics such as phase-change memory devices. Exciting further research (listed in below table) awaiting in this direction spans high-level algorithmic exploration all the way to efficient hardware design for emerging computational fabrics.

Useful Reading

Prerequisites

  • Python
  • Background in machine learning (recommended)
  • Experience with any deep learning framework such as TensorFlow or PyTorch (recommended)
  • VLSI I (recommended)


Available Projects

We are inviting applications from students to conduct their thesis (bachelor, semester, and master) or an internship project at the IBM Research lab in Zurich on this exciting new topic.

Type Project Description Workload Type
MA Code Generation for Analog In- Memory Computing SoCs Link to description algorithmic design
MA Neurosymbolic Architectures to Approach Human-like AI Link to description algorithmic design
MA Developing Efficient Models for Solving Intelligence Quotient (IQ) Test Link to description algorithmic/hardware design
MA Crytography meets in-memory computing Link to description algorithmic design and hardware experiments
MA Optimal routing for 2D Mesh-based Analog Compute-In-Memory Accelerator Architecture Link to description hardware design

Contact

Thesis will be at IBM Zurich in Rüschlikon
Hybrid AI Systems (HAS) projects
Contact (at ETH Zurich): Dr. Frank K. Gurkaynak
Contact (at IBM): Dr. Irem Boybat
Contact (at IBM): Dr. Manuel Le Gallo
Contact (at IBM): Dr. Abu Sebastian
Contact (at IBM): Dr. Abbas Rahimi
Professor: Luca Benini