text stringlengths 3 8.33k | repo stringclasses 52
values | path stringlengths 6 141 | language stringclasses 35
values | sha stringlengths 64 64 | chunk_index int32 0 273 | n_tokens int32 1 896 |
|---|---|---|---|---|---|---|
nalgorithm\n24 / 25"
},
{
"id": "lecture-06-p025",
"sourceId": "lecture-06",
"lectureId": "lecture-06",
"page": 25,
"title": "Wolpert, D. (1992).",
"text": "Feature Selection & Construction\nModel Selection and Parameter Optimisation\nEnsemble Methods\nPost-Processing of Pred... | mlpc-exam | lib/data/slide-index.json | JSON | f402ccacf2db5ca8458160410fd5226be7ea522ea2427d0acac4ae00e415aea3 | 161 | 896 |
n▶Derive the Error Back-propagation Algorithm for multi-layer ANN learning\n▶Discuss how to use ANNs for classification\n2 / 56",
"keywords": [
"networks",
"for",
"neural",
"anns",
"the",
"arti",
"cial",
"multi-layer",
"mlps",
"classi"... | mlpc-exam | lib/data/slide-index.json | JSON | e8d6ce62dbdcbb88d5a104aa1f13a9030fa02316f1406a67c78b62adaae02c18 | 162 | 896 |
x)\nThere are many algorithms for this (you aready know some ...).\nHere: focus on Logistic Regression and its generalisation, Neural Networks\n1Terminology: A Regression Model is a model that predicts one or several numeric values from a\nset of independent variables (features).\n4 / 56",
"keywords": [
"... | mlpc-exam | lib/data/slide-index.json | JSON | b496f87486227ce971b87b19e578583fb41cd5bde4640f16c5c00c926a74d367 | 163 | 896 |
for training networks with hidden layers\n5 / 56"
},
{
"id": "lecture-07-p006",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 6,
"title": "6 / 56",
"text": "Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classi... | mlpc-exam | lib/data/slide-index.json | JSON | ab8f3f5935c25f1c583a7cedac3c8b6ad959809c324c2562ff9aa50c50157597 | 164 | 896 |
And not such a\ndork all the time?\nThe Terminator: My CPU is a neural net processor; a\nlearning computer. But Skynet pre-sets\nthe switch to read-only when we’re sent\nout alone.\nSarah Connor: Doesn’t want you doing too much\nthinking, huh?\nThe Terminator: No.\n8 / 56"
},
{
"id": "lecture-07-p009",
... | mlpc-exam | lib/data/slide-index.json | JSON | 39e7eafc3f51e3356af371018e7f52e7789bdf47f44d3efdc1af776294c70ff4 | 165 | 896 |
Unit i receives inputs xj through connections from some other units\n▶Each input connection has an associated weight wij\n▶Calculations carried out by a unit i:\n1 Compute weighted sum of the inputs (“net input”): neti = P\nj wijxj\n2 Apply an activation function f to the net input: yi = f(neti)\n3 Pass output yi on (a... | mlpc-exam | lib/data/slide-index.json | JSON | 206e3541761aec3e21bf54fdcfa5939904f2969e2a58f0fff85142dfcb99c5c2 | 166 | 896 |
": "lecture-07-p012",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 12,
"title": "Learning Task:",
"text": "Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classification\nLiterature\nA Gentle Introduction: The Regression Task... | mlpc-exam | lib/data/slide-index.json | JSON | e67fd9490cfded8a168e7141f46ffd5b12c786ae75c262a001010febc0aafcb8 | 167 | 896 |
(w1xi + w0))2 is a quadratic function\nof model parameters w1 and w0\n⇒has a unique minimum.\n13 / 56"
},
{
"id": "lecture-07-p014",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 14,
"title": "General Algorithm for Approximate Solution: Iterative Gradient Descent",
... | mlpc-exam | lib/data/slide-index.json | JSON | 85ff1254f7db184edf17aff05c8d940da13d58191b6a7a20ab69f8d169dbb1e0 | 168 | 896 |
mlps",
"parameter",
"gradient",
"move",
"arti",
"cial",
"anns",
"feed-forward",
"multi-layer"
],
"retrievalText": "Neural Networks. 15 / 56. Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Class... | mlpc-exam | lib/data/slide-index.json | JSON | bcfd6adf681c42850f6ebb81c1f90818cd9e27a8d7d990643966fd8fcbac3320 | 169 | 896 |
y) > 0\n⇒prediction was too high; need to decrease slope w1 if x\nwas positive and increase it if x was negative\n▶Analogously for other cases, and for intercept w0\n16 / 56"
},
{
"id": "lecture-07-p017",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 17,
"title": "17... | mlpc-exam | lib/data/slide-index.json | JSON | d27c9bef0d745a7d45962b02355878775c1cf15abc1bc9c9629e87a2c677c221 | 170 | 896 |
data:\n⇒Might be better approximated by a curved (non-linear) function!\n19 / 56",
"keywords": [
"networks",
"neural",
"mlps",
"non-linear",
"arti",
"cial",
"anns",
"feed-forward",
"multi-layer",
"perceptrons",
"for",
... | mlpc-exam | lib/data/slide-index.json | JSON | 96ce95ec1b16490552e31791174ecddd4964413d62d462c3621909c6679cce15 | 171 | 896 |
Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classification\nLiterature\nHidden Units and Activation Functions\nMulti-layer Feed-forward Networks\nExtended network computes a\nnon-linear function:\ny = w21 tanh(w11x + w10) + w20\nRoles of the four weights in the fitting process:\n▶Bias... | mlpc-exam | lib/data/slide-index.json | JSON | 4c48855aad93ec765ae1b1e2a4b1ce945d0da270dc7101bd5759116cdda6adf1 | 172 | 896 |
of mi units, respectively\n▶an output layer of m units, to code (possibly multiple) outputs\n▶Connections: all units of a layer are connected to all units of the next layer.\nUsual (‘classical’) configuration:\n▶Only one hidden layer (sufficient for\nuniversal function approximation)\n▶Input values in range (e.g., [−1, 1... | mlpc-exam | lib/data/slide-index.json | JSON | 8d2ea8736a8e93e9513202b8410a0f1b2589cbc355ca97d717567db6ba8b3a42 | 173 | 896 |
nThe tanh activation function:\n▶Adds non-linearity to the function\ncomputed by the network\n▶Can be seen as a soft threshold\nfunction – similar to a step function,\nbut differentiable\n\u0011 Remember introductory slide:\n“biological neurons fire when their input\nexceeds a certain threshold”\nTheoretical Result (Hor... | mlpc-exam | lib/data/slide-index.json | JSON | 4e77b3a9886426a255f53d98c2f236d993c244096561d9b47745b1d4e48d3952 | 174 | 896 |
": "lecture-07",
"lectureId": "lecture-07",
"page": 26,
"title": "y2 = W T",
"text": "Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classification\nLiterature\nFeed-forward ANNs in Concise Matrix Notation\nConcise Mathematical Summary of Feed-Fo... | mlpc-exam | lib/data/slide-index.json | JSON | 26b9d91cff0878af2d842a1140f2392b8487e1ac3ae750593d0365707aaec1f3 | 175 | 896 |
nLearning Task:\n▶Given dataset D = {(xi, yi)} of training examples with input values xi and\n(numeric) target values yi\n▶Learn (find) values for the network’s parameters θ = {Wl, bl}\n(i.e., values for all the weights in the network)\n▶such that a given error function E is minimised.\nBasic Idea:\n▶Learn by Iterative ... | mlpc-exam | lib/data/slide-index.json | JSON | 55bfc67e40b646ee47c8380ece859443568e044c02a43f4b0011a2a68a55f456 | 176 | 896 |
Ancestors’: the set of units that feed into j\nSj = {k : ∃wjk}\n‘Successors’: the set of units that j feeds into.\n28 / 56"
},
{
"id": "lecture-07-p029",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 29,
"title": "i wijxi is a function of its incoming weights wij",
... | mlpc-exam | lib/data/slide-index.json | JSON | 0e88c64b5b1b6a9cebd447211dfd069785aba999e8da5b93a6ffe6d87c82170b | 177 | 896 |
\n▶For each layer l = 1, ..., L, propagate values through units at layer l (using\nthe current setting of the weights):\nyl = f(W T\nl yl−1 + bl)\n▶Obtain final output values ˆy at the output units.\n▶Store all yl at all levels for backward pass.\n30 / 56\nSS 2026:\nIgnore for Exam",
"keywords": [
"the",
... | mlpc-exam | lib/data/slide-index.json | JSON | f745e8092fab7606517e91b927dc88181e3b56df4baccd82eea976c601c13f6e | 178 | 896 |
=\n∂\n∂netj [1/2(yj −netj)2]\n=\n−(yj −netj)\n=\n−(yj −ˆyj)\n=\n(ˆyj −yj)\n3because we assume output units use identity function yi = f(neti) = neti\n31 / 56\nSS 2026:\nIgnore for Exam"
},
{
"id": "lecture-07-p032",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 32,
"... | mlpc-exam | lib/data/slide-index.json | JSON | 7e9c2cf131252710234926398afbdea891014fba2e5aab100e702e4307a3c5e4 | 179 | 896 |
=\n∂E\n∂netj\n=\nX\nk∈Sj\n∂E\n∂netk\n∂netk\n∂yj\n∂yj\n∂netj\n=\nX\nk∈Sj\nδkwjkf ′\nj(netj)\n=\nf ′\nj(netj)\nX\nk∈Sj\nδkwjk\n33 / 56\nSS 2026:\nIgnore for Exam",
"keywords": [
"for",
"the",
"netj",
"networks",
"error",
"neural",
"mlps",
"signal",
... | mlpc-exam | lib/data/slide-index.json | JSON | 70f27ce54e818cb792f76aad87f3ec702b7785a79afb772b92348edf165b7fc3 | 180 | 896 |
\nMLPs for Classification\nLiterature\nDeriving the Gradient of the Error Function\nError Backpropagation\nPutting it all together:\nThe Gradient ∆wij = ∂E/∂wij needed for learning is (see above)\n∆wij\n=\nδjyi\n=\nyif ′\nj(netj)\nX\nk∈Sj\nδkwjk\nIn Words:\nThe gradient of the error function with respect to the ith inco... | mlpc-exam | lib/data/slide-index.json | JSON | d637b5c42bc931a6b7cccf9120e1ebefb6b795165b1c27df68128152c1c2194a | 181 | 896 |
a multi-layer network/“perceptron” (MLP) with L + 1 levels\n(level 0 is input units; level L is output units)\n▶defined via parameters θ: L weight matrices Wl and bias vectors bl\n▶a training set D = {(xi, yi)}\n▶an error function E(θ, D)\n▶a learning rate η\nThe Backpropagation Training Algorithm\n▶Initialise weights a... | mlpc-exam | lib/data/slide-index.json | JSON | de3e02eb0947eccbd1dbfdfe46321d56f0208274274dedfc682a234110ae7ce0 | 182 | 896 |
:\n38 / 56",
"keywords": [
"networks",
"mlps",
"the",
"neural",
"for",
"backprop",
"output",
"arti",
"cial",
"anns",
"feed-forward",
"multi-layer"
],
"retrievalText": "Neural Networks. 38 / 56. Artificial Ne... | mlpc-exam | lib/data/slide-index.json | JSON | 27ce5ad361587446af2af0cda4ae3bb953da58d5c1879b5916f223a60f601dac | 183 | 896 |
:\n▶Learning rate = 0.3, online learning (stochastic gradient descent)\n▶Output is the ‘humpbacked’ curve\n41 / 56",
"keywords": [
"networks",
"mlps",
"neural",
"for",
"the",
"backprop",
"arti",
"cial",
"anns",
"feed-forward",
... | mlpc-exam | lib/data/slide-index.json | JSON | fdb0d9c96f5ce256492e1702fddfc4565f1cf896fec6c247559ff3991899f443 | 184 | 896 |
above):\n▶Weights are updated immediately after seeing each individual data point\n▶Also called stochastic (noisy) gradient descent: gradient for a single data\npoint can be considered noisy approximation to overall gradient G:\nIntermediate Strategy (often used in practice):\n▶Mini-batch learning\n43 / 56"
},
... | mlpc-exam | lib/data/slide-index.json | JSON | 468ba270231daeaa3eefa88dd16a0bd820ba9b7004461ff12133b5c479cd20e7 | 185 | 896 |
\n45 / 56"
},
{
"id": "lecture-07-p046",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 46,
"title": "free parameters:",
"text": "Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classification\nLiterature\nBias, V... | mlpc-exam | lib/data/slide-index.json | JSON | d562c19d23dca136920147660d1ff486ab95ff134fed04538e49ad1737c37d9f | 186 | 896 |
to fit the noise in the training points T , its error on. Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classification\nLiterature\nBias, Variance, and Overfitting Avoidance\nOverfitting Avoidance via Early Stopping\nMethod:\n▶Randomly split data D into two sets: training... | mlpc-exam | lib/data/slide-index.json | JSON | 4f0d95d05529dd0acc0e6e2639763ccb09f048a4bad1346c1024ef7abd117e52 | 187 | 896 |
linear discriminant can be learned by a\nnetwork without any hidden units\n(“simple perceptron”, linear network)\nThis kind of problem can only be solved\nwith a hidden layer of appropriate size)\n49 / 56",
"keywords": [
"network",
"for",
"networks",
"output",
"neural",
... | mlpc-exam | lib/data/slide-index.json | JSON | ddfd445fea7f05e14605784f89fbd5df0fb6b2d7b0f94178b1d7fc15afea1689 | 188 | 896 |
},
{
"id": "lecture-07-p051",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 51,
"title": "(one-hot encoding)",
"text": "Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classification\nLiterature\nThe General Case: k-... | mlpc-exam | lib/data/slide-index.json | JSON | 5ff66f45af52de64983dbe648a76845b8173055c5560459b1628c3b86aa73e71 | 189 | 896 |
perceptrons",
"for",
"classi",
"cation"
],
"retrievalText": "Neural Networks. 1 x. Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classification\nLiterature\nTowards Deep (Feed-Forward) Neural Networks\nGoing from one hidden layer ...... | mlpc-exam | lib/data/slide-index.json | JSON | 1a8be91e9de3ae9639a0a8fac397c29d03b130b5d0e3e5420613e26c8278f0f4 | 190 | 896 |
:\nx\nh1\nh2\ny\n54 / 56"
},
{
"id": "lecture-07-p055",
"sourceId": "lecture-07",
"lectureId": "lecture-07",
"page": 55,
"title": "▶basic idea of how the gradient is propagated back through the network",
"text": "Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-... | mlpc-exam | lib/data/slide-index.json | JSON | b4d330c3ab5c8df66431a08834d220e47a4d8049efaf76d2825ef61a4e520cba | 191 | 896 |
,
"mlps",
"literature"
],
"retrievalText": "Neural Networks. Introduction to Neural Networks. http://nic.schraudolph.org/teach/NNcourse/.. Artificial Neural Networks (ANNs)\nFeed-Forward Networks (Multi-layer Perceptrons, MLPs)\nMLPs for Classification\nLiterature\nLiterature\nHornik, K., Stin... | mlpc-exam | lib/data/slide-index.json | JSON | 09c4ce9d077fc32c99ef769a7c70010df3df208a3fa7c75bfc3c1b6b635e982a | 192 | 896 |
do with deep learning?\n2.\nHow does it work, as far as you’ve learned?\n3.\nHow does it work in current practice?\na.\nArchitectures\nb.\nOptimization\nc.\nRegularization\nd.\nReuse\n4.\nWhat’s the catch?\n3"
},
{
"id": "lecture-08-p004",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
... | mlpc-exam | lib/data/slide-index.json | JSON | aec191416c03afe66d5c54ea4796a42d76b3bd4d3a34b3e3966d3b48d10c83de | 193 | 896 |
m³\nfly ash\n… kg/m³\nwater\n… kg/m³\nsuperplasticizer\n… kg/m³\ncoarse aggregate\n… kg/m³\nfine aggregate\n… kg/m³\nage\n… days\ncompressive strength\n?? MPa",
"keywords": [
"age",
"aggregate",
"deep",
"nonlinear",
"regression",
"task",
"what",
"for... | mlpc-exam | lib/data/slide-index.json | JSON | 68c6490d55289af5adf02322fc93a90308b52bae433ca1da23ba27fe4b346300 | 194 | 896 |
Colorful Image Colorization, https://arxiv.org/abs/1603.08511",
"text": "11\nIMAGE COLORIZATION\nTask: Create colored image from grayscale image\nColorful Image Colorization, https://arxiv.org/abs/1603.08511",
"keywords": [
"image",
"colorization",
"colorful",
"https",
... | mlpc-exam | lib/data/slide-index.json | JSON | 788cd345e5cb724065d33e6ebab244f073e47caf52b869c2707149f533e7f737 | 195 | 896 |
/karpathy.github.io/2015/05/21/rnn-effectiveness/\nhttps://chatgpt.ch/\nfor (i = 0; i < 16; i++) {\n if (k & (1 << 1))\n pipe = (in_use & UMXTHREAD_UNCCA) +\n ((count & 0x00000000fffffff8) & 0x000000f) << 8;\n if (count == 0)\n sub(pid, ppc_md.kexec_handle, 0x20000000);\n pipe_set_bytes(i, 0);\n}\n/* Free our user pages... | mlpc-exam | lib/data/slide-index.json | JSON | b3ef2eb4e8714524decd5ba9da4f4e95c076d5638e80d12d388f88bd23c5de08 | 196 | 896 |
"
},
{
"id": "lecture-08-p016",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 16,
"title": "16",
"text": "SOLVING A TASK WITH MACHINE LEARNING\n1.\nFormalize task so its solution can be expressed as a function\n2.\nDefine model as a generic solution with free p... | mlpc-exam | lib/data/slide-index.json | JSON | 45652168a3aa0de3a8787a61f56e9a6ee194d4a645dd9652ab8228168fbc8c1e | 197 | 896 |
WITH MACHINE LEARNING\n1.\nFormalize task so its solution can be expressed as a function\n2.\nDefine model as a generic solution with free parameters\n3.\nDefine loss function measuring how bad the solution is\n4.\nOptimize model parameters to minimize loss\nY = f(X)\n18"
},
{
"id": "lecture-08-p019",
... | mlpc-exam | lib/data/slide-index.json | JSON | 9a38e1dd94efe0d72177d860e3ff063f95387629e03ecbe7cd45c879bd3b8d09 | 198 | 896 |
b2 + W2\nTσ(b1 + W1\nTx)))\nx\nb1 + W1\nTx\ny\nh1\nb2 + W2\nTh1\nh2\nb3 + W3\nTh2\n21"
},
{
"id": "lecture-08-p022",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 22,
"title": "“dense layer”",
"text": "Generic solution with free parameters, composing nonlinear ... | mlpc-exam | lib/data/slide-index.json | JSON | 8d96ed52af7bfeea2746730d0ccbbafb123dd2cbccf329ed95f21d10b2224157 | 199 | 896 |
L(θ; f)\n24"
},
{
"id": "lecture-08-p025",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 25,
"title": "25",
"text": "SOLVING A TASK WITH MACHINE LEARNING\n1.\nFormalize task so its solution can be expressed as a function\n2.\nDefine model as a generic solution ... | mlpc-exam | lib/data/slide-index.json | JSON | 7a2b96edb3651f7e1593cd6a82a10e72ef7fc0da24ffb71f248311dd96572c25 | 200 | 896 |
,
"its",
"can",
"expressed"
],
"retrievalText": "Deep Learning. l = L(θ; f, D) = 𝚺(X, T) ∈ D J(f(X; θ), T). SOLVING A TASK WITH DEEP LEARNING\n1.\nFormalize task so its solution can be expressed as a function\n2.\nDefine model as a generic solution with free parameters\n3.\nDefine l... | mlpc-exam | lib/data/slide-index.json | JSON | c27ea3752ab760cbca160f58a0ac40f036fecd21f101b26f67a92c5f13c06629 | 201 | 896 |
; θ)\nl = L(θ; f, D) = 𝚺(X, T) ∈ D J(f(X; θ), T)\nθ* = argminθ L(θ; f, D)\n29"
},
{
"id": "lecture-08-p030",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 30,
"title": "3.",
"text": "GRADIENT DESCENT\n4.\nOptimize model parameters to minimize loss\nIterative s... | mlpc-exam | lib/data/slide-index.json | JSON | 80fda6ab9504704ee8838b5dfee0fc5c064d94576920c0694931422c7bda566d | 202 | 896 |
\ninitialize θ randomly\n1.\nfind direction in which L decreases\nmost quickly: the gradient of L wrt. θ\n2.\nmove θ a bit into that direction\n3.\ngo to step 1\nθ* = argminθ L(θ; f, D)\nθ1\nθ2\nθ0 ~ N(μ=0, σ=0.01)\n32"
},
{
"id": "lecture-08-p033",
"sourceId": "lecture-08",
"lectureId": "lect... | mlpc-exam | lib/data/slide-index.json | JSON | 1fa62f527950e2758386553dbd569278fc9fb572010379b103ec62d87a766ead | 203 | 896 |
",
"text": "GRADIENT DESCENT\n4.\nOptimize model parameters to minimize loss\nIterative scheme:\n0.\ninitialize θ randomly\n1.\nfind direction in which L decreases\nmost quickly: the gradient of L wrt. θ\n2.\nmove θ a bit into that direction\n3.\ngo to step 1\nθ* = argminθ L(θ; f, D)\nθ1\nθ2\nθ0 ~ N(μ=0, σ=0.01)\... | mlpc-exam | lib/data/slide-index.json | JSON | 4cc6e3aab1fed4984712357d375aa8824901af2a04d9b5c0a6a159454bff140e | 204 | 896 |
’(θk) = -[∂L / ∂θ]θ=θk\nθk+1 = θk + η · vk\n37",
"keywords": [
"gradient",
"direction",
"deep",
"descent",
"optimize",
"parameters",
"minimize",
"loss",
"iterative",
"scheme",
"initialize",
"randomly"
],
"r... | mlpc-exam | lib/data/slide-index.json | JSON | 626d3b5e8eb3c51d286e1fe4a50a5ea23a110055b2094f6513a7142528066002 | 205 | 896 |
a function\n2.\nDefine model as a generic solution with free parameters\n3.\nDefine loss function measuring how bad the solution is\n4.\nOptimize model parameters to minimize loss\nY = f(X; θ)\nl = L(θ; f, D) = 𝚺(X, T) ∈ D J(f(X; θ), T)\nθ* = argminθ L(θ; f, D)\n39"
},
{
"id": "lecture-08-p040",
"s... | mlpc-exam | lib/data/slide-index.json | JSON | cb4bc6b3235783a60a35169679a53351f0a510bef2c9bc3073aa661b11885d53 | 206 | 896 |
lecture-08",
"lectureId": "lecture-08",
"page": 43,
"title": "… kg/m³",
"text": "EXAMPLE: NONLINEAR REGRESSION\nTask: Predict at what force a concrete cylinder bursts, depending on\n component quantities and age\nSolution form: y = f(x)\nInput x: 8-dimensional vector\nOutput y: scalar\ncement\n…... | mlpc-exam | lib/data/slide-index.json | JSON | 18106ef91390504f2b8217473cf20f4c1786551a6c1734f83e686055bb744ff8 | 207 | 896 |
,
{
"id": "lecture-08-p046",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 46,
"title": "46",
"text": "BINARY IMAGE CLASSIFICATION\nTask: Distinguish grayscale photographs of chihuahuas and blueberry\n muffins\nSolution form: y = f(X)\nInput X: matrix of gray value... | mlpc-exam | lib/data/slide-index.json | JSON | bb7493ed1454adf6c48e402677b4114f9769edb50c32c8eab0b68bf5b8a63cd2 | 208 | 896 |
with a fixed set of classes)\nSolution form: y = f(X)\nInput X: 3-tensor of RGB values\nOutput y: vector of class probabilities\nX ∈ [0,1]28×28\ny ∈ [0,1]10 ; ∑i yi = 1.0\nX ∈ [0,1]3×32×32\ny ∈ [0,1]10 ; ∑i yi = 1.0\n(1,0,0, … 0)\n(0,0,1, … 0)\n48"
},
{
"id": "lecture-08-p049",
"sourceId": "lecture-... | mlpc-exam | lib/data/slide-index.json | JSON | 86d3e6ef677e9939d542e080a94c6b335af8a8c72bc156bbdb2771134a952537 | 209 | 896 |
\n(0.392, -0.124, ...)\nx ∈ ℝ100\n52",
"keywords": [
"image",
"deep",
"generation",
"task",
"create",
"colored",
"scratch",
"possibly",
"domain-specific",
"solution",
"form",
"input"
],
"retrievalText": "De... | mlpc-exam | lib/data/slide-index.json | JSON | 78f983822b108400ce52a582118fdd43914ecd2e4c1fdfdc132077d29fd52ccb | 210 | 896 |
; θ)\n55"
},
{
"id": "lecture-08-p056",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 56,
"title": "“dense layer”",
"text": "Generic solution with free parameters, composing nonlinear functions:\nExpression can be visualized as a graph:\nMULTI-LAYER PERCEPTRON\... | mlpc-exam | lib/data/slide-index.json | JSON | 0a5a5d09314ad8db21c67804fd3b0a04fc0ca9f5470be1c37bb415164416e821 | 211 | 896 |
petal \nlength, sepal length, petal width)\n58",
"keywords": [
"the",
"sepal",
"length",
"width",
"petal",
"task",
"input",
"iris",
"example",
"scalar",
"weights",
"deep"
],
"retrievalText": "Deep Learning.... | mlpc-exam | lib/data/slide-index.json | JSON | 3fb208dc524ea59d3128bb4193bc0f59f9d7bf66350c913ca5c3033849655484 | 212 | 896 |
lectureId": "lecture-08",
"page": 61,
"title": "1",
"text": "CONVOLUTION\n61\nFormal definition:\n(X ∗ W)[a,b] = ∑i ∑j X[a + i, b + j] W[i, j]\nInformal definition:\nWe take an input tensor, X\nand a (usually smaller) kernel, W\nand slide W over X, summing\nthe elementwise product\nfor each possible p... | mlpc-exam | lib/data/slide-index.json | JSON | 6f0115994bd74eefefcb06cd6a0d147bdd97e985b9014a2f7c01eacd98e91f5d | 213 | 896 |
] = ∑i ∑j X[a + i, b + j] W[i, j]\nInformal definition:\nWe take an input tensor, X\nand a (usually smaller) kernel, W\nand slide W over X, summing\nthe elementwise product\nfor each possible position.\nCheck https://setosa.io/ev/image-kernels/ for an interactive demo.\n1\n3\n0\n2\n4\n2\n1\n0\n2\n2\n3\n2\n0\n1\n2\n0\n1... | mlpc-exam | lib/data/slide-index.json | JSON | 394db73c097af5346ae45cafca545fc2e8e92e736cddd5c05b8e0c39e2cd8748 | 214 | 896 |
\nfew learnable weights\n●\nWeights are required to be \napplicable at every position (and \nthus have to be generic enough)\nCONVOLUTIONAL LAYERS\nConvolutional layer:\nEach input is a tensor,\neach weight is a tensor,\neach output is the sum of\ninputs convolved by weights.\n65",
"keywords": [
"weights"... | mlpc-exam | lib/data/slide-index.json | JSON | 8220392fdbde13fdd01f4426950ff2f0c86c8aff85b658e71879603e3201fc01 | 215 | 896 |
-connected layers: integrate information over full input\nNot only useful for images, but any data with spatial or temporal structure \n(audio signals, volumetric scans, video, …).\nX\ny\nconv\nconv\npool\npool\ndense\ndense\n= 0.0"
},
{
"id": "lecture-08-p068",
"sourceId": "lecture-08",
"lect... | mlpc-exam | lib/data/slide-index.json | JSON | bdd8886966154790b1b87ec68fc1d67ef114e9f23e4d292643f40db474eb3895 | 216 | 896 |
",
"for",
"deep",
"dense",
"layers",
"and"
],
"retrievalText": "Deep Learning. ●. FULLY-CONVOLUTIONAL NEURAL NETWORKS\nconv\nconv\nconv\nconv\nX\nY\n=\nconv\nFully-convolutional neural networks have no dense layers (and possibly \nno pooling). Use cases include:\n●\nI... | mlpc-exam | lib/data/slide-index.json | JSON | 8f0362e5f8d888c2f6e50c2b279f4bbab80c96bb5b42f7e41b65185dd9fa00ae | 217 | 896 |
08-p073",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 73,
"title": "73",
"text": "MORE ARCHITECTURES FOR SEQUENCES\nWhat feed-forward architectures cannot do:\n●\nSequence generation:\nTake a fixed-size input, output a sequence of variable length (e.g., \ngenerating text f... | mlpc-exam | lib/data/slide-index.json | JSON | b294ef143d43203f5aa9ea3539398145a80515fd722f3fd00d0d75417e63ddcc | 218 | 896 |
let the network \nproduce multiple Yi until it produces an end-of-sequence marker.\n●\nSequence transduction: We can have two RNNs: an encoder to \nproduce a representation of the input sequence, and a decoder to \ngenerate a sequence from it.\n75\nX1\nY1\nh0\nX2\nY2\nh1\nX3\nY3\nh2\nX4\nY4\nh3\nh4",
"keywords": ... | mlpc-exam | lib/data/slide-index.json | JSON | b58a7d9aea117331c05879472d6c47db873a9da138588abe07bad3c80448bb6d | 219 | 896 |
process each Xi of a sequence by comparing it to all other Xi \n(via “self-attention”)\n●\nLike RNNs, can be used for all four sequence tasks\n●\nIn contrast to RNNs, self-attention guarantees the full sequence is \ntaken into account for each prediction\n●\nCaveat: computation requirements grow quadratically with \nse... | mlpc-exam | lib/data/slide-index.json | JSON | 33437a2f6658ab5dbcecd78b4c863deb74f1f8c920025e9f75cfcd5de3101c9b | 220 | 896 |
L(θ; f, D) = 𝚺(X, T) ∈ D J(f(X; θ), T)\nθ* = argminθ L(θ; f, D)\n80"
},
{
"id": "lecture-08-p081",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 81,
"title": "3.",
"text": "GRADIENT DESCENT\n4.\nOptimize model parameters to minimize loss\nIterative scheme:\n0.... | mlpc-exam | lib/data/slide-index.json | JSON | 638c2d4d6cc02b8999854d6457ede955e595bc6a997cfa8854888a62aca3e038 | 221 | 896 |
\nstandard recipes exist since 2010.\nThis was a major reason networks in \nthe 2000’s only had very few layers!\nθ1\nθ2\n83"
},
{
"id": "lecture-08-p084",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 84,
"title": "84",
"text": "GRADIENT DESCENT\nWhat can poss... | mlpc-exam | lib/data/slide-index.json | JSON | ab458595422dfc86bfd036be9868ad20fa40fa4f07e01e1fe2c4508c15ad71f5 | 222 | 896 |
",
"keywords": [
"small",
"deep",
"what",
"can",
"possibly",
"wrong",
"rate",
"progress",
"will",
"slow",
"regions",
"gradients"
],
"retrievalText": "Deep Learning. 87. What can possibly go wrong?\n4.\nWith... | mlpc-exam | lib/data/slide-index.json | JSON | d8490cc6068d9be2dc6786c624ac78553eb1917629793154c5f8f2f1046e09d7 | 223 | 896 |
2"
},
{
"id": "lecture-08-p091",
"sourceId": "lecture-08",
"lectureId": "lecture-08",
"page": 91,
"title": "θ2",
"text": "What can possibly go wrong?\n5.\nDifferent phases require a large \nlearning rate (to escape saddle points) \nor small rate (to hit a local optimum).\nSol... | mlpc-exam | lib/data/slide-index.json | JSON | cb8106522019fcc6fd8d7d7f746735e897327b55d5b8911ba5255efddbdb5014 | 224 | 896 |
deep",
"the",
"classical",
"sweet",
"spot",
"are",
"complex",
"they"
],
"retrievalText": "Deep Learning. the classical “sweet spot” (https://arxiv.org/abs/1812.11118). 94\nNO REGULARIZATION VIA MODEL SIZE\n●\nModern deep learning models are complex eno... | mlpc-exam | lib/data/slide-index.json | JSON | c3384eae68d1155dfef0cd511d5986c420b75f78b74758162fd43fad39b48b9b | 225 | 896 |
"augmentation",
"transform",
"for",
"any",
"ideas",
"deep",
"regularization",
"via",
"training",
"let"
],
"retrievalText": "Deep Learning. dense. REGULARIZATION VIA DATA AUGMENTATION\nData augmentation:\nTransform training data, let cla... | mlpc-exam | lib/data/slide-index.json | JSON | 3482b66aa246ef62a9b7d596ca0ba9a60ba306c0fe3178a08a7ef6a19ca36bb2 | 226 | 896 |
n5\n6\n7\n8\n9\nσ\ndense\nsoftmax\nW1 b1\nW2 b2\nW3 b3\nz\ndropout\ndropout\n100",
"keywords": [
"dropout",
"weights",
"drop",
"units",
"scale",
"not",
"dense",
"deep",
"regularization",
"via",
"effective",
"trick"
... | mlpc-exam | lib/data/slide-index.json | JSON | d049fe56e6248a9095e9392b7ae784649621b986925a8bc68ed297b261bd96d3 | 227 | 896 |
nNot enough information to constrain a complex model\n●\nTransfer learning:\n○\nTake model trained on somewhat similar task / data\nX\ny\nconv\nconv\npool\npool\ndense\ndense\n1.0\n0.0\n0.0\n0.0\n …\n0.0\n=\npizza\npasta\ncake\ncookie\n …\ndonut\n103"
},
{
"id": "lecture-08-p103",
"sourceId": "lectu... | mlpc-exam | lib/data/slide-index.json | JSON | 36789868f1149f09b5acceb4dd9595a391d08a8c618dd5e5352d965f86a01f01 | 228 | 896 |
new task (possibly only the head)\n●\nIntuition: First layers learn transformations / features which are \ngeneral enough to be useful for different tasks\nX\ny\nconv\nconv\npool\npool\ndense\ndense\n= 0.0\n106",
"keywords": [
"transfer",
"training",
"data",
"enough",
"task... | mlpc-exam | lib/data/slide-index.json | JSON | 391e2203fb4da267c5850702ab6ed5fc707b3aa1c224bc6f9effcd9fe0d351bd | 229 | 896 |
\ndense\ndense\n1.0\n0.0\n0.0\n0.0\n …\n0.0\n=\npizza\npasta\ncake\ncookie\n …\ndonut\n108",
"keywords": [
"deep",
"for",
"easier",
"access",
"most",
"software",
"packages",
"maintain",
"public",
"not",
"and",
"conv"
... | mlpc-exam | lib/data/slide-index.json | JSON | 889d01ad0e3c55f4f9e8954455662c0123f6ff8ebe04d429c03d741a38fdf8eb | 230 | 896 |
Bousquet (Google):\nDeep Learning community ≈ a huge genetic algorithm trying \ndifferent things and keeping the ones that work consistently\nphotograph: Stephen Edmonds, cc-by-sa 2.0\n112",
"keywords": [
"deep",
"and",
"alchemy",
"the",
"built",
"bridge",
"... | mlpc-exam | lib/data/slide-index.json | JSON | c486acd39e3065ae69bdc0df8bd6b69f7c643240a0dc1deda7ffd5273342ac3e | 231 | 896 |
\nhttp://colah.github.io/posts/2015-08-Understanding-LSTMs/\n●\nHints for applying deep learning to a new task:\nhttps://karpathy.github.io/2019/04/25/recipe/\nFree watching:\n●\nVideo lectures: http://course.fast.ai, \nhttps://www.youtube.com/user/ProfNandoDF/videos\n114",
"keywords": [
"http",
"... | mlpc-exam | lib/data/slide-index.json | JSON | 5282e0501f238f68779b1464a765a858160e97b52609e16e4160f6d8273ac775 | 232 | 896 |
"2/50",
"text": "UE Project Scenario: Overview\nSED System\nmicrowave oven\nrunning water\n...\n0\n5\n10\n15\n20\n2/50",
"keywords": [
"deep",
"for",
"sound",
"event",
"detection",
"project",
"scenario",
"overview",
"sed",
"syst... | mlpc-exam | lib/data/slide-index.json | JSON | 5e54bb955c7cbbafc76e93762b81c76d08278fa8a5863020f51d9160ff1072d1 | 233 | 896 |
Recap (1/3)\nWhich of the deep learning concepts that we discussed last time are particularly\nrelevant for our sound event detection task?\n■Browse through the slides from last time1 and write down the concepts\nthat you think will be relevant relevant for training a DL-based sound event\ndetection model.\n(2 min)\n1h... | mlpc-exam | lib/data/slide-index.json | JSON | ba000db350fbf8a3289e30cdc9deaa97017a8e1c0e971d6f8df0ab0cb32917d4 | 234 | 896 |
sed",
"system",
"microwave",
"oven",
"running",
"water"
],
"retrievalText": "Deep Learning for Sound Event Detection. 7/50. Formalization\nSED System\nmicrowave oven\nrunning water\n...\n0\n5\n10\n15\n20\nˆY = f(x; θ)\n7/50"
},
{
"id": "lecture-09-p012",... | mlpc-exam | lib/data/slide-index.json | JSON | 59d1b91aa2e4bc84e7dc559d3c71ee29117343db23249d4f3bf9141b2e4877ea | 235 | 896 |
",
"convolution",
"time",
"frequency",
"kernel",
"spectrogram"
],
"retrievalText": "Deep Learning for Sound Event Detection. 9/50. Convolution (1/2)\nTime\nFrequency\nKernel\nSpectrogram\n(X ∗s K)(i, j) =\nX\nm\nX\nn\nX(s · i −m, s · j −n) K(m, n)\n9/50"
},
{
... | mlpc-exam | lib/data/slide-index.json | JSON | 7e30d9b243bfa884c281d2d2ef73e249731195fec6c15124bb8eb775a163c6ab | 236 | 896 |
50"
},
{
"id": "lecture-09-p019",
"sourceId": "lecture-09",
"lectureId": "lecture-09",
"page": 19,
"title": "13/50",
"text": "Strided Convolution (2/2)\nTime\nFrequency\nKernel\nSpectrogram\n(X ∗s K)(i, j) =\nX\nm\nX\nn\nX(s · i −m, s · j −n) K(m, n)\n13/50",
"keywords"... | mlpc-exam | lib/data/slide-index.json | JSON | 616a7daa3fecbe7921e1ac83b13999500aa3a69bb456f4b8a864f4f9d939843a | 237 | 896 |
lecture-09-p023",
"sourceId": "lecture-09",
"lectureId": "lecture-09",
"page": 23,
"title": "17/50",
"text": "Convolutional Layer 2\nkernel 3\nkernel 2\nkernel 1\n17/50",
"keywords": [
"kernel",
"deep",
"for",
"sound",
"event",
"detecti... | mlpc-exam | lib/data/slide-index.json | JSON | 9509c3cea17a53c9145734b41a65ba21410d3db78a91f533397dfaf0ce2bb8fb | 238 | 896 |
ncomplete transformation:\nhl+1 = F(hl),\nInstead: just learn a tiny correction:\nhl+1 = hl + F(hl).\nx\ny\nBatchNorm\nReLU\nConv\nBatchNorm\nReLU\nConv\n+\n19/50",
"keywords": [
"batchnorm",
"relu",
"conv",
"deep",
"for",
"sound",
"event",
"detectio... | mlpc-exam | lib/data/slide-index.json | JSON | d5e979069001f1644512394a229db082720f3f90ba5dbda106b062242f216604 | 239 | 896 |
": 32,
"title": "23/50",
"text": "RNN\nRNN Cell\nRNN Cell\n23/50",
"keywords": [
"rnn",
"cell",
"deep",
"for",
"sound",
"event",
"detection"
],
"retrievalText": "Deep Learning for Sound Event Detection. 23/50. RNN\nRNN Cell\nRNN Cell\... | mlpc-exam | lib/data/slide-index.json | JSON | fa4b133c0cabde3e9de4a9190093b033c5a45d819d0ef5247e2c787b1bf3aa64 | 240 | 896 |
id": "lecture-09-p037",
"sourceId": "lecture-09",
"lectureId": "lecture-09",
"page": 37,
"title": "27/50",
"text": "📋\nTask: Concept Check\nRate the following answer as true, mostly true, or false.\n27/50",
"keywords": [
"deep",
"for",
"sound",
"event... | mlpc-exam | lib/data/slide-index.json | JSON | 4cd38e17d8f03c6dd31561796ba349eb5946e4c8606c1fdb7a7e1600236e1092 | 241 | 896 |
ny = 1 :\nlog(y)\ny = 0 :\nlog(1\ny)\n30/50",
"keywords": [
"log",
"loss",
"microwave",
"oven",
"ground",
"truth",
"predicted",
"deep",
"for",
"sound",
"event",
"detection"
],
"retrievalText": "Deep Learnin... | mlpc-exam | lib/data/slide-index.json | JSON | 7df6d6ee5eb06416f2cda0e5f305c55185c15f9580b1d4d2d06d7a6259325969 | 242 | 896 |
"sourceId": "lecture-09",
"lectureId": "lecture-09",
"page": 44,
"title": "■. . .",
"text": "Ok, but where do we start?\n■Architecture\n□kernel, channels, stride, depth, activations, batch norm\n□hidden size, number of layer, bi-directions\n■Optimization\n□learning rate, momentum, weight decay\n... | mlpc-exam | lib/data/slide-index.json | JSON | 5f888d774181977dedeefbcb05f2cb57283635abbef4d02a346333f031b9247a | 243 | 896 |
n63 × 63\nRB 4\n128\n3 × 3, 1 × 1\n2 × 2, s = 2\n16 × 377\n87 × 87\nRB 5\n256\n1 × 1, 1 × 1\n–\n8 × 188\n87 × 87\nRB 6\n256\n1 × 1, 1 × 1\n–\n8 × 188\n87 × 87\nRB 7\n256\n1 × 1, 1 × 1\n–\n8 × 188\n87 × 87\nRB 8\n256\n1 × 1, 1 × 1\n–\n8 × 188\n87 × 87\nBiGRU 1\n256\n2048 × 188\nBiGRU 2\n256\n512 × 188\nPredict\n15\n15 ×... | mlpc-exam | lib/data/slide-index.json | JSON | 3d0a540372ed27a88a5c44def185ab38dbe6e24b900aa3800fbda49c98ac79c9 | 244 | 896 |
n■Evaluate all possible combinations\nRandom Search\n■Define ranges\n■Sample randomly\n36/50",
"keywords": [
"search",
"deep",
"for",
"sound",
"event",
"detection",
"systematically",
"tune",
"hyperparameters",
"goal",
"find",
... | mlpc-exam | lib/data/slide-index.json | JSON | e31342c3283115681602d92e8171e86a063402a500ab338a4cfc8d82e013a637 | 245 | 896 |
n38/50"
},
{
"id": "lecture-09-p053",
"sourceId": "lecture-09",
"lectureId": "lecture-09",
"page": 53,
"title": "0 = 10",
"text": "Learning Rate Scheduler\n0\n10\n20\n30\n40\nEpoch\n0.0000\n0.0005\n0.0010\nLearning Rate\n0 = 10\n3\nmin = 10\n5\nHyperparameters\n■lr_decay_star... | mlpc-exam | lib/data/slide-index.json | JSON | 368383ad6a0af8344a4bf116833a4f9d85b828da6f21b119e08edc96c7115952 | 246 | 896 |
": "lecture-09",
"lectureId": "lecture-09",
"page": 56,
"title": "42/50",
"text": "Data Augmentation\nGoal: Generate additional training examples using domain-specific\ntransformations.\n(˜x, ˜Y) = T((x1, Y1), . . . , (xk, Yk), z)\n■(xi, Yi) ∼D are training examples\n■z denotes optional auxiliary... | mlpc-exam | lib/data/slide-index.json | JSON | 75b419ee46a8686dadace34992f170fe0d51ad396d1057560004024dd0897dbf | 247 | 896 |
0\n2.5\n5.0\n7.5\n10.0\n12.5\n15.0\n17.5\n0\n100\nHyperparameters\n■p: probability of applying the augmentation\n■db_range: minimum and maximum gain (dB)\n■n_band_min: min number of frequency bands\n■n_band_max: max number of frequency bands\n44/50"
},
{
"id": "lecture-09-p059",
"sourceId": "lecture... | mlpc-exam | lib/data/slide-index.json | JSON | 3f9cdfcfb99d1b49deb656c0b6287920ee0ff27a9eaf079306485b897e5b226b | 248 | 896 |
architecture for SED,\n■training it on our dataset, and\n■tuning basic hyperparameters\n■augmentation techniques\n47/50",
"keywords": [
"for",
"basic",
"deep",
"sound",
"event",
"detection",
"things",
"covered",
"today",
"architecture... | mlpc-exam | lib/data/slide-index.json | JSON | 1ebbf7f08ba0a8ec73012689c5c61ba467a5882065b6823a2c1211443733b23a | 249 | 896 |
.\n■I’m skeptical about W; does it actually work?\n50/50",
"keywords": [
"was",
"the",
"concept",
"you",
"never",
"heard",
"deep",
"for",
"sound",
"event",
"detection",
"task"
],
"retrievalText": "Deep Lear... | mlpc-exam | lib/data/slide-index.json | JSON | c2b6fd22a4d8a0513d4a824868c2a8736cb9dac3b7c39fe0c6a541b91552ef0c | 250 | 896 |
Length Principle\n3 Statistical Data Modelling\nGaussian Mixture Models (GMMs) and the E-M Algorithm\nRelation to Clustering, and to Density Estimation\n3 / 55",
"keywords": [
"clustering",
"and",
"the",
"relation",
"density",
"estimation",
"motivation",
... | mlpc-exam | lib/data/slide-index.json | JSON | c3c2bd6814993b2a32ae833673250dac52f0c295f9391fe63464ddaf491966ac | 251 | 896 |
"seem",
"three"
],
"retrievalText": "Unsupervised Learning. 5 / 55. Motivation\nClustering\nStatistical Data Modelling\nGoal 1: Find Structure in the Data\n“There seem to be three distinct groups of points in the data, with one being farther apart\n(i.e., more different) from the other two.”\n5 / 55... | mlpc-exam | lib/data/slide-index.json | JSON | 46c16f9a158db374d74b3bbc50abe3a99f5570a8307be3c1b918ceeca38ce2fa | 252 | 896 |
▶co-variances Σ1 = Σ3 =\n\u0012\n1\n0\n0\n1\n\u0013\n, Σ2 =\n\u0012\n1\n1\n1\n2\n\u0013\n▶and prior probabilities (relative weights) π1 ≈1/4, π2 ≈1/2, π3 ≈1/4.”\n8 / 55",
"keywords": [
"data",
"unsupervised",
"motivation",
"clustering",
"statistical",
"modelling",
... | mlpc-exam | lib/data/slide-index.json | JSON | 0c6d95554bc7412b4e59b9c51be05144aaf2eb75049928f461e711443ef4b3cf | 253 | 896 |
lecture-10",
"page": 10,
"title": "1",
"text": "Motivation\nClustering\nStatistical Data Modelling\nA classical clustering algorithm: k-means\nk-means Clustering: The Objective Function\nWhat is a good partitioning of the data into disjoint clusters?\n▶Want instances within a cluster to be similar (cl... | mlpc-exam | lib/data/slide-index.json | JSON | 03f172d2dbd7f6187e0d726f7577d9d7eaa16f4db8a255eca00ece59918299bb | 254 | 896 |
\n▶Parameter k (number of clusters)\nInitialise: Start with k randomly chosen cluster centers ci ∈X\nLoop:\n1 Assign each xi ∈D to the cluster (center) nearest to it\n2 For each cluster Ci: compute new cluster center ci as the center (mean) of\nall points currently in Ci:\nci ←\n1\n|Ci|\nX\nxj∈Ci\nxj\n3 Compute quality... | mlpc-exam | lib/data/slide-index.json | JSON | 29e656291f326b3d8703f777f508bdda9763316fc72c37afedfaacb25e64598a | 255 | 896 |
,
"simple",
"example"
],
"retrievalText": "Unsupervised Learning. 15 / 55. Motivation\nClustering\nStatistical Data Modelling\nA classical clustering algorithm: k-means\nk-means Clustering: A Simple Example\n15 / 55"
},
{
"id": "lecture-10-p016",
"sourceId": "lecture-10",... | mlpc-exam | lib/data/slide-index.json | JSON | 502e5e2fa115c2fb5b365342dec71c98a21ea3272feb490fdba99bf09f0038c4 | 256 | 896 |
Motivation\nClustering\nStatistical Data Modelling\nA classical clustering algorithm: k-means\nk-means Clustering: A Simple Example\n20 / 55",
"keywords": [
"clustering",
"k-means",
"unsupervised",
"motivation",
"statistical",
"data",
"modelling",
"c... | mlpc-exam | lib/data/slide-index.json | JSON | 1173ec04483732d8c1e4790221bfc23df60f57446828af5d1792e2fe40597974 | 257 | 896 |
retrievalText": "Unsupervised Learning. 24 / 55. Motivation\nClustering\nStatistical Data Modelling\nA classical clustering algorithm: k-means\nk-means Clustering: Yet Another Episode\n24 / 55"
},
{
"id": "lecture-10-p025",
"sourceId": "lecture-10",
"lectureId": "lecture-10",
"page": 25,... | mlpc-exam | lib/data/slide-index.json | JSON | 48b81e3d4f03d5651421db2a47df0207c214bc147ea4f4e0e6532f3bb4fdd168 | 258 | 896 |
text": "Motivation\nClustering\nStatistical Data Modelling\nModel Selection: What is the best k?\nModel Selection: Strategies for Finding a Good Setting for k\nScenario:\n▶Run k-means algorithm with a range of different values k\n▶Evaluate the resulting clusterings via some quality measure\n▶Choose the k and correspond... | mlpc-exam | lib/data/slide-index.json | JSON | 9875b9e2df660a35bd0caf3b3f3b6715cea7ecc2cb18a65c70f4404b47909f6d | 259 | 896 |
\nBIC and the Minimum Message Length Principle\nModel Selection via the Bayes Information Criterion (BIC)\nExcursus:1\n▶The BIC arises from a Bayesian analysis of a model M’s probability, given a\ndata set D (which we want to maximise):\nP(M | D) = P(D | M)P(M)\nP(D)\n∝P(D | M)P(M) ∝log P(D | M) + log P(M)\nDifferent f... | mlpc-exam | lib/data/slide-index.json | JSON | 6d2bd21170c75c986e87f803de19564c0171212357a17c07f3028ec8f7d92326 | 260 | 896 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.