Only Truth Matters

1. Introduction

https://sccn.ucsd.edu/~mgrivich/Synchronization.html

Goal: properly synchronize events in the EEG experiments and validate the sychronization.

Why? verify that the timing of all data points is known with sufficient precision. Some key points such as the playing of a sound, the display of an image.

Error

• constant offset
• drift: as time passes, the difference increase
• jitter:

2. General Principles

1. When possible, use hardware synchronization
2. streaming, in general, 50ms is easily achievable, and 1ms accuracy is achievable with effort
3. behavior must be validated

3. Audio Synchronization

Playing audio is not reliable with respect to timing. Latencies of 100+/-50 ms are common.

0.1. Introduction

Problem of the Listener: Extract meaning from the acoustic signal

• involves the recognition of words
• identifying the perceptual cues that listeners use

0.2. Cues

• aspirated and unaspirated stops in Mandarin differ in VOT
• Are listeners senstive to these differences?
• Most test: manipulate the acoustic property synthetically and see if perceptual response is affected

Properties of cues:

• cues combine to distinguish sounds
• individual cues can vary in strength

1. Categorical

Listeners only have access to category labels, so stimuli can only be distinguished if they are identified as belonging to different categories

Why significant?

1. Physics

Sound is a pressure wave, pressure fluctuations move through space,but each air particle moves only a small distance.

While using a microphone to capture sounds, the physics behind it is AIR PRESSURE, so the waveform files (.wav/.mp3…) will have negative and positive value!

2. Spectrums

The spectrums of a sound plays a center role in determining its quality

we can represent the sound in frequency plot

• the quality of a vowel depends on the shape of its spectrum
• the peaks are called formants
• the quality depends primarily on the first three formant

1. 1 History

• 1875, observed from exposed brains
• 1914, photographed experimentally induced seizures
• 1934, “human brain waves”,

2. 2 Why measure EEG

Greatest advantage: temporal resolution, EEG can determine electrical activity in different brain regions

3. 3 Physics

Necessary conditions: Aligned neurons and synchronous activity

Acquisition:

• electrode caps
• conductive jelly
• ruler
• EEG amplifier, laptop

4. 4 Data collection

1. Three-Pass Approach

Key idea: read a paper up tp 3 passes, each pass accomplishes specific goals

1.1. The First Pass

quick scan, decide whether you need to do any more passes, take about 5-10 minutes

1. read the title and abstract, introduction carefully
2. read section headings
3. read the conclusions
4. glance the references

After reading, you should get five Cs

• category
• context
• correctness
• contributions
• clarity

1.2. The Second Pass

read the paper with greater care, but ignore the details such as proofs, take up to 1 hour,

1. look carefully at the figures, diagrams
2. mark relevant unread references

1. Introduction

5 Topics:

• Syntax: the structure of language,

• subj, verb, object

• English: SVO; Japanese: SOV

• create a syntax tree

• Semantics: the meaning of language

• Morphology: the structure of words

• Birds, Pets, the difference between s

• Phonetics: peoperties of sounds

• Phonogy: patterns of sounds

• pay/bay/ebay, bay sounds like pay

2. Syntax 1

2.1. Word order

English word order is SVO (subject verb object)

2.2. Grammaticality

• prescriptive: what should be said/writen
• descriptive: observation about what is said

1. Overview

Neuroscience-Inspired Artificial Intelligence, a paper released by DeepMind in Cell

Why neuroscience is important for AI?

• inspiration
• validation

But in practical, biological plausibility is just a guide, not a strict requirement.

What can we do? Marr and Poggio stated that there are 3 layers of analysis

1. top level: the goal of the system
2. mid level: the process and computations, also called algorithmic level
3. bottom level: how to implement the system physically

What DeepMind do is focusing the top 2 level of the analysis

2. Some History

2.1. Deep Learning

1. intro

AI have significant benefits like

• Faster processing speed
• no emotional judge

But AI system cannot be value free (introduced by developer)

2. What is value

• subjective preferences
• standards for evaluation (judge the extent), in this sense, not only human have value
• in this sense, there is a plurality of different standards

3. System

No system is perfect. All system involve tradeoffs, trade off reflect value

Value in system is ineliminable. In 3 phase

• Problem Identification
• different ways to address problem
• different definition
• Design
• data collection, privacy
• which data to train, feature or not?
• algorithmic fairness
• Implementation

1. Overview

This article is a note I wrote while reading the post in nature

The are mainly 6 fields in the cognitive science community

• Psychology
• Linguistics
• Artificial Intelligence
• Anthropology
• Philosophy
• Neuroscience

Cognitive science has a short history if we check out the creation of the flagship journal in this fiel—Cognitive Science—is in1977.

In the post, author use bibliometric indicators and socio-institutional indicators to analyze the development of cognitive science

2. The Challenge

The difficulties from the beginning was achieving interdisciplinary

Howard Gardner wrote an influential book Mind‘s New Science. In the book, Howard characterized some essnetial original features of the cognitive science

• it’s necessary to speak about mental representations and separate from biological
• “the faith that central to any understanding of the human mind is the electronic computer.” — ragard brain as a device like computer

1. Introduction

Control theory provides a systetmatic approach to design feedback loops that are stable and settle quickly to their steady state values.

In many computing systems, control theory is used widely in problem such as adjusting scheduling priorites, memory allocations, and network bandwith allocation and flow control and TCP/IP

2. Control Theoty Fundamentals

The magic of feedback: create a system that perform well from components that perform poorly.

​ — Karl Astrom

2.1. Notions

Because we are talking about computer system, so the time is discrete. Here are some symbols

• time: $k$
• reference input: $r(k)$, the desired value of measured output
• control error: $e(k)$, the difference between reference input and the measured output
• control output $u(k)$, the output of controller. This is a parameter in the target system
• disturbance input $d(k)$, any changes that affects the way in which the $u(k)$ influences the measured output
• measured output $y(k)$
• noise input $n(k)$ changes the measured output produced by the target system

2.2. Purpose

• regulatory control: ensure the measured output is close to the reference input
• disturbance rejection
• optimization: obtain the best value of the measured output