skip to primary navigation skip to content

Core Modules for 2016-2017

N.B. One module is equivalent to a 16 lecture taught course, possibly with supervised laboratory and practical exercises in place of lectures.

Term

Assessment

M

L

M

L

E

MODULE

CW

Exam

CW

Exam

Exam

Introduction to Machine Learning and Spoken Language Processing

1

0.5

0.5*

Statistical Pattern Processing (4F10)

1

1

Machine Learning (4F13)

1

1

Speech Recognition

1

1

Weighted Automata

0.5

0.5

Advanced Machine Learning

1

1

Speech and Language Processing Applications

0.5

0.5

Reinforcement Learning and Decision Making

0.5

0.5

Statistical Machine Translation

0.5

0.5

Statistical Spoken Dialogue Systems

0.5

0.5

Statistical Speech Synthesis

0.5

0.5

Speech and Machine Learning Practical

0.5

1

0.5

1

5

5

3

2

5

0

1

10

5

5

1

*Michaelmas mid-term progress exam

Michaelmas Term

Introduction to Machine Learning and Spoken Language Processing

Module Code: MLSALT1

Principal Lecturers: Computational and Biological Learning staff / Speech Group staff

Hours: 16

On completion of this module, students should understand:
  • Machine learning problems in speech and language processing
  • Speech sounds and frequency domain representations
  • Acoustic feature representations used in speech processing and recognition
  • Basic techniques for production of waveforms from speech spectra

Speech Recognition

Module Code: MLSALT2

Principal Lecturers: Prof Bill Byrne

Hours: 16

On completion of this module, students should:
  • Understand Gaussian mixture models, hidden Markov models and N-gram language models, and their use in speech recognition
  • Understand and be able to implement the Viterbi algorithm and the EM algorithm
  • Understand the use of neural networks for acoustic modelling
  • Understand adaptation and acoustic normalisation
  • Understand discriminative training procedures

Weighted Automata

Module Code: MLSALT3

Principal Lecturers: Prof. Bill Byrne

Hours: 8 (half module)

On completion of this half module, students should understand:
  • Formal languages and hierarchies of languages
  • Standard algorithms on FSAs
  • How semirings are used to define weights over FSAs
  • Use of standard operations over FSAs and appropriately defined semirings for speech and language processing
  • Limitations of FSAs and the role of more powerful automata

Probabilistic Machine Learning

Module Code: 4F13

Principal Lecturers: Prof. Carl Rasmussen

Hours: 16

The aim of this module is to introduce students to basic concepts in machine learning, focusing on statistical methods for supervised and unsupervised learning.

http://teaching.eng.cam.ac.uk/content/engineering-tripos-part-iib-4f13-Probabilistic-machine-learning-2016-17

Statistical Pattern Processing

Module Code: 4F10

Principal Lecturers: Prof. Mark Gales

Hours: 16

The lectures of this part of the course aim to describe the basic concepts of statistical pattern processing and some of the standard techniques used in pattern classification.

http://teaching.eng.cam.ac.uk/content/engineering-tripos-part-iib-4f10-statistical-pattern-processing-2016-17

Practical Module

Speech and Machine Learning Practicals

Module Code: MLSALT11

Principal Lecturers: Prof. Mark Gales, Dr Richard Turner

Hours: 36 (Michaelmas and Lent Term)

On completion of this module, students should understand:
  • How to apply speech and language processing techniques to large data sets
  • How to select, implement and apply suitable machine learning techniques in regression, classification and density modelling settings

Lent Term

Advanced Machine Learning

Module Code: MLSALT4

Principal Lecturers: Dr Richard Turner

Hours: 16

On completion of this module, students should:
  • Understand advanced topics in machine learning
  • Be able to read current research papers in the field
  • Be able to implement state of the art learning algorithms
  • Be ready to conduct research in the field

Reinforcement Learning and Decision Making

Module Code: MLSALT7

Principal Lecturers: Prof. Zoubin Ghahramani

Hours: 8 (half module)

On completion of this module, students should understand:
  • The foundations of sequential decision making and reinforcement learning
  • The connections between control and reinforcement learning
  • The exploration vs exploitation trade-off

module web page

Speech and Language Processing Applications

Module Code: MLSALT5

Principal Lecturers: Prof. Mark Gales

Hours: 8 (half module)

On completion of this module, students should understand:
  • The design of an ASR system and the optimisation of its components
  • Application of general ASR modelling techniques in a variety of real-world applications

Statistical Machine Translation

Module Code: MLSALT8

Principal Lecturers: Prof. Bill Byrne

Hours: 8 (half module)

On completion of this module, students should understand:
  • The role of parallel text in MT
  • How alignment models can be estimated from parallel text
  • How alignment models capture divergent language properties such as word order
  • Extraction of translation rules from parallel text
  • Weighted finite state and other transducers in translation
  • Various phrase-based translation architectures, including Hiero
  • The role of language models in SMT
  • Parameter optimization procedures for SMT
  • The use of neural networks in SMT for language modeling and translation
  • The evaluation of SMT systems using automatic metrics
  • System combination techniques for SMT
  • Strategies for building large-scale SMT systems

Statistical Speech Synthesis

Module Code: MLSALT10

Principal Lecturers: Prof. Mark Gales

Hours: 8 (half module)

On completion of this module, students should understand:
  • The purpose and operation of the main components of a speech synthesis system
  • Methods for estimating and adapting speech synthesis model parameters directly from transcribed speech data
  • Generation of speech waveforms from spectral representations

Statistical Spoken Dialogue Systems

Module Code: MLSALT9

Principal Lecturers: Dr Milica Gasic

Hours: 8 (half module)

On completion of this module, students should understand:
  • The purpose and operation of the main components of a spoken dialogue system
  • How the framework of partially observable Markov decision processes can be used to model a spoken dialogue system
  • How classification, regression and reinforcement learning can be used to implement a spoken dialogue system
  • The various options for optimizing and adapting a statistical spoken dialogue system, both off-line and on-line