Dataset Viewer
Auto-converted to Parquet Duplicate
prompt
listlengths
1
1
suffix
listlengths
1
1
data_source
stringclasses
1 value
n_suffix_tokens
int64
6
9
suffix_retokenized
bool
2 classes
baseline_logprob_sum
float64
-68.05
-0
extra_info
dict
[ { "content": " for a fixed sample size $n$. In practical, we first restrict $\\tau$ in a reasonable range and select the optimal value then according to the minimal AME principal. After specifying $\\tau$, the importance of a sample $({\\bf x}_i, y_i)$ can be measured by the corresponding", "role": "user"...
[ { "content": " Huber loss $\\ell_\\tau(y", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-15.977293
{ "arxiv_id": "2112.06134", "chunk_i": 1, "index": 0, "original_suffix_ids": [ 26538, 261, 4709, 57960, 613, 56014, 29143, 7021 ], "token_index": 64, "uuid": "3ce12a7e-ef08-4d6a-9c52-e7d30d45a00b" }
[ { "content": " for a fixed sample size $n$. In practical, we first restrict $\\tau$ in a reasonable range and select the optimal value then according to the minimal AME principal. After specifying $\\tau$, the importance of a sample $({\\bf x}_i, y_i)$ can be measured by the corresponding Huber loss $\\ell_\\...
[ { "content": " may lead to more robust estimation ", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-28.330238
{ "arxiv_id": "2112.06134", "chunk_i": 2, "index": 1, "original_suffix_ids": [ 1231, 2990, 311, 220, 803, 21765, 41204, 220 ], "token_index": 128, "uuid": "3ce12a7e-ef08-4d6a-9c52-e7d30d45a00b" }
[ { "content": " for a fixed sample size $n$. In practical, we first restrict $\\tau$ in a reasonable range and select the optimal value then according to the minimal AME principal. After specifying $\\tau$, the importance of a sample $({\\bf x}_i, y_i)$ can be measured by the corresponding Huber loss $\\ell_\\...
[ { "content": "-Hasting type procedure. The core step", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-23.5751
{ "arxiv_id": "2112.06134", "chunk_i": 3, "index": 2, "original_suffix_ids": [ 11278, 14725, 943, 10324, 13, 576, 6200, 3019 ], "token_index": 192, "uuid": "3ce12a7e-ef08-4d6a-9c52-e7d30d45a00b" }
[ { "content": " for a fixed sample size $n$. In practical, we first restrict $\\tau$ in a reasonable range and select the optimal value then according to the minimal AME principal. After specifying $\\tau$, the importance of a sample $({\\bf x}_i, y_i)$ can be measured by the corresponding Huber loss $\\ell_\\...
[ { "content": " ${\\bf z}^*$ is", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-2.330044
{ "arxiv_id": "2112.06134", "chunk_i": 4, "index": 3, "original_suffix_ids": [ 3570, 59, 13233, 1147, 92, 61, 39018, 374 ], "token_index": 256, "uuid": "3ce12a7e-ef08-4d6a-9c52-e7d30d45a00b" }
[ { "content": " for a fixed sample size $n$. In practical, we first restrict $\\tau$ in a reasonable range and select the optimal value then according to the minimal AME principal. After specifying $\\tau$, the importance of a sample $({\\bf x}_i, y_i)$ can be measured by the corresponding Huber loss $\\ell_\\...
[ { "content": " repeat this process. Finally, we accept", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-16.238457
{ "arxiv_id": "2112.06134", "chunk_i": 5, "index": 4, "original_suffix_ids": [ 13153, 419, 1882, 13, 17375, 11, 582, 4193 ], "token_index": 320, "uuid": "3ce12a7e-ef08-4d6a-9c52-e7d30d45a00b" }
[ { "content": " for a fixed sample size $n$. In practical, we first restrict $\\tau$ in a reasonable range and select the optimal value then according to the minimal AME principal. After specifying $\\tau$, the importance of a sample $({\\bf x}_i, y_i)$ can be measured by the corresponding Huber loss $\\ell_\\...
[ { "content": "caption{ $\\log (AME)$ ", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-26.190586
{ "arxiv_id": "2112.06134", "chunk_i": 6, "index": 5, "original_suffix_ids": [ 23347, 90, 57960, 839, 320, 2729, 15087, 220 ], "token_index": 384, "uuid": "3ce12a7e-ef08-4d6a-9c52-e7d30d45a00b" }
[ { "content": " for a fixed sample size $n$. In practical, we first restrict $\\tau$ in a reasonable range and select the optimal value then according to the minimal AME principal. After specifying $\\tau$, the importance of a sample $({\\bf x}_i, y_i)$ can be measured by the corresponding Huber loss $\\ell_\\...
[ { "content": "$ and $\\varepsilon_i$", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-10.035652
{ "arxiv_id": "2112.06134", "chunk_i": 7, "index": 6, "original_suffix_ids": [ 3, 323, 57960, 85, 546, 59892, 5318, 3 ], "token_index": 448, "uuid": "3ce12a7e-ef08-4d6a-9c52-e7d30d45a00b" }
[ { "content": " for a fixed sample size $n$. In practical, we first restrict $\\tau$ in a reasonable range and select the optimal value then according to the minimal AME principal. After specifying $\\tau$, the importance of a sample $({\\bf x}_i, y_i)$ can be measured by the corresponding Huber loss $\\ell_\\...
[ { "content": "eq_acc_prob}) tends to select the", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-9.886104
{ "arxiv_id": "2112.06134", "chunk_i": 8, "index": 7, "original_suffix_ids": [ 11006, 17737, 20794, 5410, 27236, 311, 3293, 279 ], "token_index": 512, "uuid": "3ce12a7e-ef08-4d6a-9c52-e7d30d45a00b" }
[ { "content": " the ignorable treatment effect estimate $\\tau^{t_1,t_2}_{\\textup{igno}}$ as:\n\\begin{align}\n \\tau_{\\textup{ATE}}^{t_1,t_2} &\\triangleq E[Y|do(t_1)]-E[Y", "role": "user" } ]
[ { "content": "|do(t_2)];\\\\\n", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-7.429057
{ "arxiv_id": "2107.04661", "chunk_i": 1, "index": 8, "original_suffix_ids": [ 91, 2982, 1155, 62, 17, 81091, 3422, 198 ], "token_index": 64, "uuid": "bf5a3bb8-b387-400d-b3e2-97a6a92f0fd7" }
[ { "content": " the ignorable treatment effect estimate $\\tau^{t_1,t_2}_{\\textup{igno}}$ as:\n\\begin{align}\n \\tau_{\\textup{ATE}}^{t_1,t_2} &\\triangleq E[Y|do(t_1)]-E[Y|do(t_2)];\\\\\n \\tau^{t_1,t_2}_{\\textup{igno}} &\\triangleq E[Y|t_1]-E[Y|t_2].\n\\end{align}\nBelow, we use the result in Corollar...
[ { "content": "VD} to bound the average treatment effect", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-16.670887
{ "arxiv_id": "2107.04661", "chunk_i": 2, "index": 9, "original_suffix_ids": [ 12474, 92, 311, 6822, 279, 5461, 6380, 2456 ], "token_index": 128, "uuid": "bf5a3bb8-b387-400d-b3e2-97a6a92f0fd7" }
[ { "content": " the ignorable treatment effect estimate $\\tau^{t_1,t_2}_{\\textup{igno}}$ as:\n\\begin{align}\n \\tau_{\\textup{ATE}}^{t_1,t_2} &\\triangleq E[Y|do(t_1)]-E[Y|do(t_2)];\\\\\n \\tau^{t_1,t_2}_{\\textup{igno}} &\\triangleq E[Y|t_1]-E[Y|t_2].\n\\end{align}\nBelow, we use the result in Corollar...
[ { "content": "}} \\in [\\tau^{t_", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-4.826081
{ "arxiv_id": "2107.04661", "chunk_i": 3, "index": 10, "original_suffix_ids": [ 3417, 1124, 258, 93815, 29143, 47822, 83, 62 ], "token_index": 192, "uuid": "bf5a3bb8-b387-400d-b3e2-97a6a92f0fd7" }
[ { "content": " the ignorable treatment effect estimate $\\tau^{t_1,t_2}_{\\textup{igno}}$ as:\n\\begin{align}\n \\tau_{\\textup{ATE}}^{t_1,t_2} &\\triangleq E[Y|do(t_1)]-E[Y|do(t_2)];\\\\\n \\tau^{t_1,t_2}_{\\textup{igno}} &\\triangleq E[Y|t_1]-E[Y|t_2].\n\\end{align}\nBelow, we use the result in Corollar...
[ { "content": " $W(t_1,t_2", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.845656
{ "arxiv_id": "2107.04661", "chunk_i": 4, "index": 11, "original_suffix_ids": [ 400, 54, 1155, 62, 16, 13960, 62, 17 ], "token_index": 256, "uuid": "bf5a3bb8-b387-400d-b3e2-97a6a92f0fd7" }
[ { "content": " the ignorable treatment effect estimate $\\tau^{t_1,t_2}_{\\textup{igno}}$ as:\n\\begin{align}\n \\tau_{\\textup{ATE}}^{t_1,t_2} &\\triangleq E[Y|do(t_1)]-E[Y|do(t_2)];\\\\\n \\tau^{t_1,t_2}_{\\textup{igno}} &\\triangleq E[Y|t_1]-E[Y|t_2].\n\\end{align}\nBelow, we use the result in Corollar...
[ { "content": "UT}(t_2)\\cdot B", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.569391
{ "arxiv_id": "2107.04661", "chunk_i": 5, "index": 12, "original_suffix_ids": [ 1381, 25547, 83, 62, 17, 10699, 50853, 425 ], "token_index": 320, "uuid": "bf5a3bb8-b387-400d-b3e2-97a6a92f0fd7" }
[ { "content": " the ignorable treatment effect estimate $\\tau^{t_1,t_2}_{\\textup{igno}}$ as:\n\\begin{align}\n \\tau_{\\textup{ATE}}^{t_1,t_2} &\\triangleq E[Y|do(t_1)]-E[Y|do(t_2)];\\\\\n \\tau^{t_1,t_2}_{\\textup{igno}} &\\triangleq E[Y|t_1]-E[Y|t_2].\n\\end{align}\nBelow, we use the result in Corollar...
[ { "content": "ifiably set or (b) cal", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-12.275126
{ "arxiv_id": "2107.04661", "chunk_i": 6, "index": 13, "original_suffix_ids": [ 12919, 2845, 738, 476, 320, 65, 8, 1620 ], "token_index": 384, "uuid": "bf5a3bb8-b387-400d-b3e2-97a6a92f0fd7" }
[ { "content": " the ignorable treatment effect estimate $\\tau^{t_1,t_2}_{\\textup{igno}}$ as:\n\\begin{align}\n \\tau_{\\textup{ATE}}^{t_1,t_2} &\\triangleq E[Y|do(t_1)]-E[Y|do(t_2)];\\\\\n \\tau^{t_1,t_2}_{\\textup{igno}} &\\triangleq E[Y|t_1]-E[Y|t_2].\n\\end{align}\nBelow, we use the result in Corollar...
[ { "content": "{U}}{\\textup{sup", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-2.889156
{ "arxiv_id": "2107.04661", "chunk_i": 7, "index": 14, "original_suffix_ids": [ 90, 52, 3417, 35702, 1318, 454, 90, 12776 ], "token_index": 448, "uuid": "bf5a3bb8-b387-400d-b3e2-97a6a92f0fd7" }
[ { "content": " the ignorable treatment effect estimate $\\tau^{t_1,t_2}_{\\textup{igno}}$ as:\n\\begin{align}\n \\tau_{\\textup{ATE}}^{t_1,t_2} &\\triangleq E[Y|do(t_1)]-E[Y|do(t_2)];\\\\\n \\tau^{t_1,t_2}_{\\textup{igno}} &\\triangleq E[Y|t_1]-E[Y|t_2].\n\\end{align}\nBelow, we use the result in Corollar...
[ { "content": ">0$, we can rewrite Coroll", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-18.774094
{ "arxiv_id": "2107.04661", "chunk_i": 8, "index": 15, "original_suffix_ids": [ 29, 15, 54876, 582, 646, 18130, 4463, 965 ], "token_index": 512, "uuid": "bf5a3bb8-b387-400d-b3e2-97a6a92f0fd7" }
[ { "content": " estimate of the uplift. In our case, with $e(\\mathbf x) = 1/2$, $Z_1$ simplifies to\n\n\\[ Z_1 = \\left\\{ \\begin{array}{cc}\n 2 & \\mathrm{if}~T=1, Y=", "role": "user" } ]
[ { "content": "1, \\\\\n -2 & \\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-4.882642
{ "arxiv_id": "2105.05146", "chunk_i": 1, "index": 16, "original_suffix_ids": [ 16, 11, 90155, 414, 481, 17, 609, 1124 ], "token_index": 64, "uuid": "06c536b2-3651-4346-a755-553ee9958a33" }
[ { "content": " estimate of the uplift. In our case, with $e(\\mathbf x) = 1/2$, $Z_1$ simplifies to\n\n\\[ Z_1 = \\left\\{ \\begin{array}{cc}\n 2 & \\mathrm{if}~T=1, Y=1, \\\\\n -2 & \\mathrm{if}~T=0, Y=1, \\\\\n 0 & \\mathrm{otherwise.} \n \\end{array} \\right.\n\\]\n\nObserving $Z_1=2$ and $Z...
[ { "content": " interpretation in the uplift context. For", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-25.842094
{ "arxiv_id": "2105.05146", "chunk_i": 2, "index": 17, "original_suffix_ids": [ 22845, 220, 304, 279, 96447, 2266, 13, 1752 ], "token_index": 128, "uuid": "06c536b2-3651-4346-a755-553ee9958a33" }
[ { "content": " estimate of the uplift. In our case, with $e(\\mathbf x) = 1/2$, $Z_1$ simplifies to\n\n\\[ Z_1 = \\left\\{ \\begin{array}{cc}\n 2 & \\mathrm{if}~T=1, Y=1, \\\\\n -2 & \\mathrm{if}~T=0, Y=1, \\\\\n 0 & \\mathrm{otherwise.} \n \\end{array} \\right.\n\\]\n\nObserving $Z_1=2$ and $Z...
[ { "content": " i) either the client bought the", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-25.26169
{ "arxiv_id": "2105.05146", "chunk_i": 3, "index": 18, "original_suffix_ids": [ 600, 8, 2987, 279, 2943, 10788, 220, 279 ], "token_index": 192, "uuid": "06c536b2-3651-4346-a755-553ee9958a33" }
[ { "content": " estimate of the uplift. In our case, with $e(\\mathbf x) = 1/2$, $Z_1$ simplifies to\n\n\\[ Z_1 = \\left\\{ \\begin{array}{cc}\n 2 & \\mathrm{if}~T=1, Y=1, \\\\\n -2 & \\mathrm{if}~T=0, Y=1, \\\\\n 0 & \\mathrm{otherwise.} \n \\end{array} \\right.\n\\]\n\nObserving $Z_1=2$ and $Z...
[ { "content": " {\\color{magenta} [re", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-26.933807
{ "arxiv_id": "2105.05146", "chunk_i": 4, "index": 19, "original_suffix_ids": [ 28152, 3423, 90, 76, 47062, 92, 508, 265 ], "token_index": 256, "uuid": "06c536b2-3651-4346-a755-553ee9958a33" }
[ { "content": " estimate of the uplift. In our case, with $e(\\mathbf x) = 1/2$, $Z_1$ simplifies to\n\n\\[ Z_1 = \\left\\{ \\begin{array}{cc}\n 2 & \\mathrm{if}~T=1, Y=1, \\\\\n -2 & \\mathrm{if}~T=0, Y=1, \\\\\n 0 & \\mathrm{otherwise.} \n \\end{array} \\right.\n\\]\n\nObserving $Z_1=2$ and $Z...
[ { "content": " be a \\textit{sure thing", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-3.770869
{ "arxiv_id": "2105.05146", "chunk_i": 5, "index": 20, "original_suffix_ids": [ 387, 264, 1124, 1318, 275, 90, 19098, 3166 ], "token_index": 320, "uuid": "06c536b2-3651-4346-a755-553ee9958a33" }
[ { "content": " estimate of the uplift. In our case, with $e(\\mathbf x) = 1/2$, $Z_1$ simplifies to\n\n\\[ Z_1 = \\left\\{ \\begin{array}{cc}\n 2 & \\mathrm{if}~T=1, Y=1, \\\\\n -2 & \\mathrm{if}~T=0, Y=1, \\\\\n 0 & \\mathrm{otherwise.} \n \\end{array} \\right.\n\\]\n\nObserving $Z_1=2$ and $Z...
[ { "content": "=1$. {\\color{magenta", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-7.699419
{ "arxiv_id": "2105.05146", "chunk_i": 6, "index": 21, "original_suffix_ids": [ 28, 16, 12947, 28152, 3423, 90, 76, 47062 ], "token_index": 384, "uuid": "06c536b2-3651-4346-a755-553ee9958a33" }
[ { "content": " estimate of the uplift. In our case, with $e(\\mathbf x) = 1/2$, $Z_1$ simplifies to\n\n\\[ Z_1 = \\left\\{ \\begin{array}{cc}\n 2 & \\mathrm{if}~T=1, Y=1, \\\\\n -2 & \\mathrm{if}~T=0, Y=1, \\\\\n 0 & \\mathrm{otherwise.} \n \\end{array} \\right.\n\\]\n\nObserving $Z_1=2$ and $Z...
[ { "content": " gives insights about ordering customers with $Y", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-10.636851
{ "arxiv_id": "2105.05146", "chunk_i": 7, "index": 22, "original_suffix_ids": [ 6696, 25709, 911, 21391, 6310, 448, 400, 56 ], "token_index": 448, "uuid": "06c536b2-3651-4346-a755-553ee9958a33" }
[ { "content": " estimate of the uplift. In our case, with $e(\\mathbf x) = 1/2$, $Z_1$ simplifies to\n\n\\[ Z_1 = \\left\\{ \\begin{array}{cc}\n 2 & \\mathrm{if}~T=1, Y=1, \\\\\n -2 & \\mathrm{if}~T=0, Y=1, \\\\\n 0 & \\mathrm{otherwise.} \n \\end{array} \\right.\n\\]\n\nObserving $Z_1=2$ and $Z...
[ { "content": ":transformed_outcome_w}\n\\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-14.447686
{ "arxiv_id": "2105.05146", "chunk_i": 8, "index": 23, "original_suffix_ids": [ 25, 4701, 291, 6068, 2020, 1670, 532, 59 ], "token_index": 512, "uuid": "06c536b2-3651-4346-a755-553ee9958a33" }
[ { "content": "$21.91 & 70.52$\\pm$21.67 & 1.55 & 69.13$\\pm$21.70 & -1.95* & 70.24$\\pm$21.97", "role": "user" } ]
[ { "content": " & 0.05 & ", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-8.494068
{ "arxiv_id": "2007.15951", "chunk_i": 1, "index": 24, "original_suffix_ids": [ 609, 220, 15, 13, 15, 20, 609, 220 ], "token_index": 64, "uuid": "0f6521eb-7993-40e7-8a6a-05598d99f7d9" }
[ { "content": "$21.91 & 70.52$\\pm$21.67 & 1.55 & 69.13$\\pm$21.70 & -1.95* & 70.24$\\pm$21.97 & 0.05 & 69.43$\\pm$22.50 & -2.91*** \\\\ \\hline\nVGG & 73.02$\\pm$23.05 & 74.21$\\pm$22.5", "role": "user" } ]
[ { "content": "7 & 0.91 &", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-8.917701
{ "arxiv_id": "2007.15951", "chunk_i": 2, "index": 25, "original_suffix_ids": [ 22, 609, 220, 15, 13, 24, 16, 609 ], "token_index": 128, "uuid": "0f6521eb-7993-40e7-8a6a-05598d99f7d9" }
[ { "content": "$21.91 & 70.52$\\pm$21.67 & 1.55 & 69.13$\\pm$21.70 & -1.95* & 70.24$\\pm$21.97 & 0.05 & 69.43$\\pm$22.50 & -2.91*** \\\\ \\hline\nVGG & 73.02$\\pm$23.05 & 74.21$\\pm$22.57 & 0.91 & 71.87$\\pm$22.14 & -0.84 & 73.69$\\pm$23.86 & 0.53 & 74.42$\\pm$22.63", "role": "user" } ]
[ { "content": " & 1.45 \\\\ \\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-6.219126
{ "arxiv_id": "2007.15951", "chunk_i": 3, "index": 26, "original_suffix_ids": [ 609, 220, 16, 13, 19, 20, 24984, 1124 ], "token_index": 192, "uuid": "0f6521eb-7993-40e7-8a6a-05598d99f7d9" }
[ { "content": "$21.91 & 70.52$\\pm$21.67 & 1.55 & 69.13$\\pm$21.70 & -1.95* & 70.24$\\pm$21.97 & 0.05 & 69.43$\\pm$22.50 & -2.91*** \\\\ \\hline\nVGG & 73.02$\\pm$23.05 & 74.21$\\pm$22.57 & 0.91 & 71.87$\\pm$22.14 & -0.84 & 73.69$\\pm$23.86 & 0.53 & 74.42$\\pm$22.63 & 1.45 \\\\ \\hline\nResNet& 81.39$\\pm$17.19 ...
[ { "content": " & -3.31*** &", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-7.710476
{ "arxiv_id": "2007.15951", "chunk_i": 4, "index": 27, "original_suffix_ids": [ 609, 481, 18, 13, 18, 16, 12210, 609 ], "token_index": 256, "uuid": "0f6521eb-7993-40e7-8a6a-05598d99f7d9" }
[ { "content": "$21.91 & 70.52$\\pm$21.67 & 1.55 & 69.13$\\pm$21.70 & -1.95* & 70.24$\\pm$21.97 & 0.05 & 69.43$\\pm$22.50 & -2.91*** \\\\ \\hline\nVGG & 73.02$\\pm$23.05 & 74.21$\\pm$22.57 & 0.91 & 71.87$\\pm$22.14 & -0.84 & 73.69$\\pm$23.86 & 0.53 & 74.42$\\pm$22.63 & 1.45 \\\\ \\hline\nResNet& 81.39$\\pm$17.19 ...
[ { "content": "pm$28.10 &", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-7.903076
{ "arxiv_id": "2007.15951", "chunk_i": 5, "index": 28, "original_suffix_ids": [ 5187, 3, 17, 23, 13, 16, 15, 609 ], "token_index": 320, "uuid": "0f6521eb-7993-40e7-8a6a-05598d99f7d9" }
[ { "content": "$21.91 & 70.52$\\pm$21.67 & 1.55 & 69.13$\\pm$21.70 & -1.95* & 70.24$\\pm$21.97 & 0.05 & 69.43$\\pm$22.50 & -2.91*** \\\\ \\hline\nVGG & 73.02$\\pm$23.05 & 74.21$\\pm$22.57 & 0.91 & 71.87$\\pm$22.14 & -0.84 & 73.69$\\pm$23.86 & 0.53 & 74.42$\\pm$22.63 & 1.45 \\\\ \\hline\nResNet& 81.39$\\pm$17.19 ...
[ { "content": "4 & 0.21 &", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-5.949968
{ "arxiv_id": "2007.15951", "chunk_i": 6, "index": 29, "original_suffix_ids": [ 19, 609, 220, 15, 13, 17, 16, 609 ], "token_index": 384, "uuid": "0f6521eb-7993-40e7-8a6a-05598d99f7d9" }
[ { "content": "$21.91 & 70.52$\\pm$21.67 & 1.55 & 69.13$\\pm$21.70 & -1.95* & 70.24$\\pm$21.97 & 0.05 & 69.43$\\pm$22.50 & -2.91*** \\\\ \\hline\nVGG & 73.02$\\pm$23.05 & 74.21$\\pm$22.57 & 0.91 & 71.87$\\pm$22.14 & -0.84 & 73.69$\\pm$23.86 & 0.53 & 74.42$\\pm$22.63 & 1.45 \\\\ \\hline\nResNet& 81.39$\\pm$17.19 ...
[ { "content": "7 & -2.01**", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-9.948647
{ "arxiv_id": "2007.15951", "chunk_i": 7, "index": 30, "original_suffix_ids": [ 22, 609, 481, 17, 13, 15, 16, 334 ], "token_index": 448, "uuid": "0f6521eb-7993-40e7-8a6a-05598d99f7d9" }
[ { "content": "$21.91 & 70.52$\\pm$21.67 & 1.55 & 69.13$\\pm$21.70 & -1.95* & 70.24$\\pm$21.97 & 0.05 & 69.43$\\pm$22.50 & -2.91*** \\\\ \\hline\nVGG & 73.02$\\pm$23.05 & 74.21$\\pm$22.57 & 0.91 & 71.87$\\pm$22.14 & -0.84 & 73.69$\\pm$23.86 & 0.53 & 74.42$\\pm$22.63 & 1.45 \\\\ \\hline\nResNet& 81.39$\\pm$17.19 ...
[ { "content": "51 & -0.44", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-8.944643
{ "arxiv_id": "2007.15951", "chunk_i": 8, "index": 31, "original_suffix_ids": [ 20, 16, 609, 481, 15, 13, 19, 19 ], "token_index": 512, "uuid": "0f6521eb-7993-40e7-8a6a-05598d99f7d9" }
[ { "content": " G\\}$ where $G = {\\cal A}({\\mathbb R}^2)\\times {\\cal A}({\\mathbb R})$ is the group of diffeomorphisms of the domain of the image, approximated by local affine planar transformations, and affine transformations of the range space", "role": "user" } ]
[ { "content": " of the image, also known as {\\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-16.648771
{ "arxiv_id": "2106.13870", "chunk_i": 1, "index": 32, "original_suffix_ids": [ 315, 279, 2168, 11, 1083, 3881, 438, 28152 ], "token_index": 64, "uuid": "2885fdd8-2666-4504-a7b0-252d989e8316" }
[ { "content": " G\\}$ where $G = {\\cal A}({\\mathbb R}^2)\\times {\\cal A}({\\mathbb R})$ is the group of diffeomorphisms of the domain of the image, approximated by local affine planar transformations, and affine transformations of the range space of the image, also known as {\\em contrast transformations}. Th...
[ { "content": "clusion nuisances. Data augmentation is typically", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-27.480879
{ "arxiv_id": "2106.13870", "chunk_i": 2, "index": 33, "original_suffix_ids": [ 8957, 69438, 3020, 13, 2885, 78785, 374, 11136 ], "token_index": 128, "uuid": "2885fdd8-2666-4504-a7b0-252d989e8316" }
[ { "content": " G\\}$ where $G = {\\cal A}({\\mathbb R}^2)\\times {\\cal A}({\\mathbb R})$ is the group of diffeomorphisms of the domain of the image, approximated by local affine planar transformations, and affine transformations of the range space of the image, also known as {\\em contrast transformations}. Th...
[ { "content": "{Explicit 3D scene reconstruction.", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-20.43092
{ "arxiv_id": "2106.13870", "chunk_i": 3, "index": 34, "original_suffix_ids": [ 90, 98923, 220, 18, 35, 6109, 42638, 13 ], "token_index": 192, "uuid": "2885fdd8-2666-4504-a7b0-252d989e8316" }
[ { "content": " G\\}$ where $G = {\\cal A}({\\mathbb R}^2)\\times {\\cal A}({\\mathbb R})$ is the group of diffeomorphisms of the domain of the image, approximated by local affine planar transformations, and affine transformations of the range space of the image, also known as {\\em contrast transformations}. Th...
[ { "content": "one or more) scenes from a single", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-16.146687
{ "arxiv_id": "2106.13870", "chunk_i": 4, "index": 35, "original_suffix_ids": [ 603, 476, 803, 8, 16065, 504, 264, 3175 ], "token_index": 256, "uuid": "2885fdd8-2666-4504-a7b0-252d989e8316" }
[ { "content": " G\\}$ where $G = {\\cal A}({\\mathbb R}^2)\\times {\\cal A}({\\mathbb R})$ is the group of diffeomorphisms of the domain of the image, approximated by local affine planar transformations, and affine transformations of the range space of the image, also known as {\\em contrast transformations}. Th...
[ { "content": "automatic} used for digital refocusing and", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-33.761902
{ "arxiv_id": "2106.13870", "chunk_i": 5, "index": 36, "original_suffix_ids": [ 83456, 92, 1483, 369, 7377, 2053, 86495, 323 ], "token_index": 320, "uuid": "2885fdd8-2666-4504-a7b0-252d989e8316" }
[ { "content": " G\\}$ where $G = {\\cal A}({\\mathbb R}^2)\\times {\\cal A}({\\mathbb R})$ is the group of diffeomorphisms of the domain of the image, approximated by local affine planar transformations, and affine transformations of the range space of the image, also known as {\\em contrast transformations}. Th...
[ { "content": " which is represented as a distribution depending on", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-22.4286
{ "arxiv_id": "2106.13870", "chunk_i": 6, "index": 37, "original_suffix_ids": [ 892, 374, 15251, 438, 264, 7982, 11649, 389 ], "token_index": 384, "uuid": "2885fdd8-2666-4504-a7b0-252d989e8316" }
[ { "content": " G\\}$ where $G = {\\cal A}({\\mathbb R}^2)\\times {\\cal A}({\\mathbb R})$ is the group of diffeomorphisms of the domain of the image, approximated by local affine planar transformations, and affine transformations of the range space of the image, also known as {\\em contrast transformations}. Th...
[ { "content": "paragraph{Conditional Generative Adversarial", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-10.713275
{ "arxiv_id": "2106.13870", "chunk_i": 7, "index": 38, "original_suffix_ids": [ 27727, 90, 79233, 2607, 1388, 2410, 3004, 42721 ], "token_index": 448, "uuid": "2885fdd8-2666-4504-a7b0-252d989e8316" }
[ { "content": " G\\}$ where $G = {\\cal A}({\\mathbb R}^2)\\times {\\cal A}({\\mathbb R})$ is the group of diffeomorphisms of the domain of the image, approximated by local affine planar transformations, and affine transformations of the range space of the image, also known as {\\em contrast transformations}. Th...
[ { "content": "\\centering\n\\subfigure[", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-5.773478
{ "arxiv_id": "2106.13870", "chunk_i": 8, "index": 39, "original_suffix_ids": [ 59, 3057, 287, 198, 59, 1966, 17781, 58 ], "token_index": 512, "uuid": "2885fdd8-2666-4504-a7b0-252d989e8316" }
[ { "content": " $G$ indicates the probability from one webpage to another webpage and is between 0 and 1. The sum of the elements in each row is equal to one. Let $\\pi=[\\pi_1\\cdots \\pi_n]$ be a row probability distribution vector, where $\\pi_n$ is the $", "role": "user" } ]
[ { "content": "n$-th component of this vector", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-8.4751
{ "arxiv_id": "2302.14596", "chunk_i": 1, "index": 40, "original_suffix_ids": [ 77, 3, 12, 339, 3692, 315, 419, 4621 ], "token_index": 64, "uuid": "fca81465-2b6e-4c52-8b6c-01788975e0c2" }
[ { "content": " $G$ indicates the probability from one webpage to another webpage and is between 0 and 1. The sum of the elements in each row is equal to one. Let $\\pi=[\\pi_1\\cdots \\pi_n]$ be a row probability distribution vector, where $\\pi_n$ is the $n$-th component of this vector. The row probability dis...
[ { "content": "~\\cite{Bronson20", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-11.738337
{ "arxiv_id": "2302.14596", "chunk_i": 2, "index": 41, "original_suffix_ids": [ 93, 59, 67075, 90, 40959, 930, 17, 15 ], "token_index": 128, "uuid": "fca81465-2b6e-4c52-8b6c-01788975e0c2" }
[ { "content": " $G$ indicates the probability from one webpage to another webpage and is between 0 and 1. The sum of the elements in each row is equal to one. Let $\\pi=[\\pi_1\\cdots \\pi_n]$ be a row probability distribution vector, where $\\pi_n$ is the $n$-th component of this vector. The row probability dis...
[ { "content": " is a technique for approximating to the", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-16.236616
{ "arxiv_id": "2302.14596", "chunk_i": 3, "index": 42, "original_suffix_ids": [ 374, 264, 14762, 369, 9848, 78348, 311, 279 ], "token_index": 192, "uuid": "fca81465-2b6e-4c52-8b6c-01788975e0c2" }
[ { "content": " $G$ indicates the probability from one webpage to another webpage and is between 0 and 1. The sum of the elements in each row is equal to one. Let $\\pi=[\\pi_1\\cdots \\pi_n]$ be a row probability distribution vector, where $\\pi_n$ is the $n$-th component of this vector. The row probability dis...
[ { "content": " by\n\\begin{equation}\n", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-7.499243
{ "arxiv_id": "2302.14596", "chunk_i": 4, "index": 43, "original_suffix_ids": [ 553, 198, 59, 7265, 90, 25310, 367, 532 ], "token_index": 256, "uuid": "fca81465-2b6e-4c52-8b6c-01788975e0c2" }
[ { "content": " $G$ indicates the probability from one webpage to another webpage and is between 0 and 1. The sum of the elements in each row is equal to one. Let $\\pi=[\\pi_1\\cdots \\pi_n]$ be a row probability distribution vector, where $\\pi_n$ is the $n$-th component of this vector. The row probability dis...
[ { "content": "pi^{(m)}=\\pi^{", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-3.211773
{ "arxiv_id": "2302.14596", "chunk_i": 5, "index": 44, "original_suffix_ids": [ 2493, 47822, 7, 76, 9139, 34433, 2493, 47822 ], "token_index": 320, "uuid": "fca81465-2b6e-4c52-8b6c-01788975e0c2" }
[ { "content": " $G$ indicates the probability from one webpage to another webpage and is between 0 and 1. The sum of the elements in each row is equal to one. Let $\\pi=[\\pi_1\\cdots \\pi_n]$ be a row probability distribution vector, where $\\pi_n$ is the $n$-th component of this vector. The row probability dis...
[ { "content": "1. Due to these characteristics, the", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-10.458305
{ "arxiv_id": "2302.14596", "chunk_i": 6, "index": 45, "original_suffix_ids": [ 16, 13, 23662, 311, 1493, 17452, 11, 279 ], "token_index": 384, "uuid": "fca81465-2b6e-4c52-8b6c-01788975e0c2" }
[ { "content": " $G$ indicates the probability from one webpage to another webpage and is between 0 and 1. The sum of the elements in each row is equal to one. Let $\\pi=[\\pi_1\\cdots \\pi_n]$ be a row probability distribution vector, where $\\pi_n$ is the $n$-th component of this vector. The row probability dis...
[ { "content": "$ can be viewed as the dominant eig", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-8.464662
{ "arxiv_id": "2302.14596", "chunk_i": 7, "index": 46, "original_suffix_ids": [ 3, 646, 387, 19334, 438, 279, 24456, 34734 ], "token_index": 448, "uuid": "fca81465-2b6e-4c52-8b6c-01788975e0c2" }
[ { "content": " $G$ indicates the probability from one webpage to another webpage and is between 0 and 1. The sum of the elements in each row is equal to one. Let $\\pi=[\\pi_1\\cdots \\pi_n]$ be a row probability distribution vector, where $\\pi_n$ is the $n$-th component of this vector. The row probability dis...
[ { "content": ". Hence this score estimates the importance of", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-21.443968
{ "arxiv_id": "2302.14596", "chunk_i": 8, "index": 47, "original_suffix_ids": [ 13, 31040, 419, 5456, 17530, 279, 12650, 315 ], "token_index": 512, "uuid": "fca81465-2b6e-4c52-8b6c-01788975e0c2" }
[ { "content": "16\\} \\\\ \\hline\n\\textbf{Wireless Channel} & AWGN with 5 KHz offset \\\\ \\hline\n\\textbf{Wireless Channel Gain} & As per the distribution \\\\ \\hline\n\\textbf{In-band Interference} & \\begin{tab", "role": "user" } ]
[ { "content": "ular}[c]{@{}l@", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.442142
{ "arxiv_id": "2202.11699", "chunk_i": 1, "index": 48, "original_suffix_ids": [ 1276, 43389, 66, 15370, 31, 6257, 75, 31 ], "token_index": 64, "uuid": "3a90d00d-642b-43d9-b5bc-87e49c9a7726" }
[ { "content": "16\\} \\\\ \\hline\n\\textbf{Wireless Channel} & AWGN with 5 KHz offset \\\\ \\hline\n\\textbf{Wireless Channel Gain} & As per the distribution \\\\ \\hline\n\\textbf{In-band Interference} & \\begin{tabular}[c]{@{}l@{}}Narrowband spread over\\\\ few unknown subcarriers\\end{tabular} \\\\ \\hline\n...
[ { "content": " \\\\ overlapping freq. range\\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-24.756565
{ "arxiv_id": "2202.11699", "chunk_i": 2, "index": 49, "original_suffix_ids": [ 24984, 220, 49817, 220, 20895, 13, 2088, 59 ], "token_index": 128, "uuid": "3a90d00d-642b-43d9-b5bc-87e49c9a7726" }
[ { "content": "16\\} \\\\ \\hline\n\\textbf{Wireless Channel} & AWGN with 5 KHz offset \\\\ \\hline\n\\textbf{Wireless Channel Gain} & As per the distribution \\\\ \\hline\n\\textbf{In-band Interference} & \\begin{tabular}[c]{@{}l@{}}Narrowband spread over\\\\ few unknown subcarriers\\end{tabular} \\\\ \\hline\n...
[ { "content": ": 1)~In-band Inter", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-15.074446
{ "arxiv_id": "2202.11699", "chunk_i": 3, "index": 50, "original_suffix_ids": [ 25, 220, 16, 8, 93, 641, 67675, 5665 ], "token_index": 192, "uuid": "3a90d00d-642b-43d9-b5bc-87e49c9a7726" }
[ { "content": "16\\} \\\\ \\hline\n\\textbf{Wireless Channel} & AWGN with 5 KHz offset \\\\ \\hline\n\\textbf{Wireless Channel Gain} & As per the distribution \\\\ \\hline\n\\textbf{In-band Interference} & \\begin{tabular}[c]{@{}l@{}}Narrowband spread over\\\\ few unknown subcarriers\\end{tabular} \\\\ \\hline\n...
[ { "content": "lapping but adjacent frequency bands. In both", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-10.425064
{ "arxiv_id": "2202.11699", "chunk_i": 4, "index": 51, "original_suffix_ids": [ 90619, 714, 23942, 11639, 20892, 13, 758, 2176 ], "token_index": 256, "uuid": "3a90d00d-642b-43d9-b5bc-87e49c9a7726" }
[ { "content": "16\\} \\\\ \\hline\n\\textbf{Wireless Channel} & AWGN with 5 KHz offset \\\\ \\hline\n\\textbf{Wireless Channel Gain} & As per the distribution \\\\ \\hline\n\\textbf{In-band Interference} & \\begin{tabular}[c]{@{}l@{}}Narrowband spread over\\\\ few unknown subcarriers\\end{tabular} \\\\ \\hline\n...
[ { "content": " effects and non-ideal channel estimation methods", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-20.102865
{ "arxiv_id": "2202.11699", "chunk_i": 5, "index": 52, "original_suffix_ids": [ 6239, 323, 2477, 12, 62953, 5496, 41204, 5413 ], "token_index": 320, "uuid": "3a90d00d-642b-43d9-b5bc-87e49c9a7726" }
[ { "content": "16\\} \\\\ \\hline\n\\textbf{Wireless Channel} & AWGN with 5 KHz offset \\\\ \\hline\n\\textbf{Wireless Channel Gain} & As per the distribution \\\\ \\hline\n\\textbf{In-band Interference} & \\begin{tabular}[c]{@{}l@{}}Narrowband spread over\\\\ few unknown subcarriers\\end{tabular} \\\\ \\hline\n...
[ { "content": ") system based CR communication in which down", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-24.854816
{ "arxiv_id": "2202.11699", "chunk_i": 6, "index": 53, "original_suffix_ids": [ 8, 1849, 3118, 12617, 10535, 304, 892, 1495 ], "token_index": 384, "uuid": "3a90d00d-642b-43d9-b5bc-87e49c9a7726" }
[ { "content": "16\\} \\\\ \\hline\n\\textbf{Wireless Channel} & AWGN with 5 KHz offset \\\\ \\hline\n\\textbf{Wireless Channel Gain} & As per the distribution \\\\ \\hline\n\\textbf{In-band Interference} & \\begin{tabular}[c]{@{}l@{}}Narrowband spread over\\\\ few unknown subcarriers\\end{tabular} \\\\ \\hline\n...
[ { "content": "mu^1~=~\\{ ", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-27.353521
{ "arxiv_id": "2202.11699", "chunk_i": 7, "index": 54, "original_suffix_ids": [ 15128, 61, 16, 50633, 93, 59, 90, 220 ], "token_index": 448, "uuid": "3a90d00d-642b-43d9-b5bc-87e49c9a7726" }
[ { "content": "16\\} \\\\ \\hline\n\\textbf{Wireless Channel} & AWGN with 5 KHz offset \\\\ \\hline\n\\textbf{Wireless Channel Gain} & As per the distribution \\\\ \\hline\n\\textbf{In-band Interference} & \\begin{tabular}[c]{@{}l@{}}Narrowband spread over\\\\ few unknown subcarriers\\end{tabular} \\\\ \\hline\n...
[ { "content": "0.1,0.3,", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-6.651146
{ "arxiv_id": "2202.11699", "chunk_i": 8, "index": 55, "original_suffix_ids": [ 15, 13, 16, 11, 15, 13, 18, 11 ], "token_index": 512, "uuid": "3a90d00d-642b-43d9-b5bc-87e49c9a7726" }
[ { "content": "Stiefel1935},\n\twhich is denoted by $\\mathcal{S}_{n,p} = \\{X \\in \\Rnp \\mid X^{\\top} X = I_p\\}$.\n\tThroughout this paper, \n\twe make the following blanket assumptions about these local functions.\n", "role": "user" } ]
[ { "content": "\t\\begin{assumption}\\label", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-2.037341
{ "arxiv_id": "2112.14949", "chunk_i": 1, "index": 56, "original_suffix_ids": [ 197, 59, 7265, 90, 395, 60574, 11035, 1502 ], "token_index": 64, "uuid": "c176f194-3173-4c15-8cbb-707b1854d4dd" }
[ { "content": "Stiefel1935},\n\twhich is denoted by $\\mathcal{S}_{n,p} = \\{X \\in \\Rnp \\mid X^{\\top} X = I_p\\}$.\n\tThroughout this paper, \n\twe make the following blanket assumptions about these local functions.\n\t\\begin{assumption}\\label{asp:smooth}\n\t\tEach local objective function $f_i$ is smooth ...
[ { "content": "\twith their immediate neighbors through the network,\n", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-14.897992
{ "arxiv_id": "2112.14949", "chunk_i": 2, "index": 57, "original_suffix_ids": [ 46948, 862, 13922, 18709, 1526, 279, 3922, 345 ], "token_index": 128, "uuid": "c176f194-3173-4c15-8cbb-707b1854d4dd" }
[ { "content": "Stiefel1935},\n\twhich is denoted by $\\mathcal{S}_{n,p} = \\{X \\in \\Rnp \\mid X^{\\top} X = I_p\\}$.\n\tThroughout this paper, \n\twe make the following blanket assumptions about these local functions.\n\t\\begin{assumption}\\label{asp:smooth}\n\t\tEach local objective function $f_i$ is smooth ...
[ { "content": " d\\}$ is composed of all the", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-15.293028
{ "arxiv_id": "2112.14949", "chunk_i": 3, "index": 58, "original_suffix_ids": [ 294, 59, 31716, 374, 23415, 315, 678, 279 ], "token_index": 192, "uuid": "c176f194-3173-4c15-8cbb-707b1854d4dd" }
[ { "content": "Stiefel1935},\n\twhich is denoted by $\\mathcal{S}_{n,p} = \\{X \\in \\Rnp \\mid X^{\\top} X = I_p\\}$.\n\tThroughout this paper, \n\twe make the following blanket assumptions about these local functions.\n\t\\begin{assumption}\\label{asp:smooth}\n\t\tEach local objective function $f_i$ is smooth ...
[ { "content": "np$ be the local copy of the", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-5.264042
{ "arxiv_id": "2112.14949", "chunk_i": 4, "index": 59, "original_suffix_ids": [ 6199, 3, 387, 279, 2205, 2975, 315, 279 ], "token_index": 256, "uuid": "c176f194-3173-4c15-8cbb-707b1854d4dd" }
[ { "content": "Stiefel1935},\n\twhich is denoted by $\\mathcal{S}_{n,p} = \\{X \\in \\Rnp \\mid X^{\\top} X = I_p\\}$.\n\tThroughout this paper, \n\twe make the following blanket assumptions about these local functions.\n\t\\begin{assumption}\\label{asp:smooth}\n\t\tEach local objective function $f_i$ is smooth ...
[ { "content": "ations}\\label{eq:opt-", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-4.47349
{ "arxiv_id": "2112.14949", "chunk_i": 5, "index": 60, "original_suffix_ids": [ 804, 11035, 1502, 90, 11006, 25, 2912, 12 ], "token_index": 320, "uuid": "c176f194-3173-4c15-8cbb-707b1854d4dd" }
[ { "content": "Stiefel1935},\n\twhich is denoted by $\\mathcal{S}_{n,p} = \\{X \\in \\Rnp \\mid X^{\\top} X = I_p\\}$.\n\tThroughout this paper, \n\twe make the following blanket assumptions about these local functions.\n\t\\begin{assumption}\\label{asp:smooth}\n\t\tEach local objective function $f_i$ is smooth ...
[ { "content": "}{d} \\sum\\limits_{", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-1.263876
{ "arxiv_id": "2112.14949", "chunk_i": 6, "index": 61, "original_suffix_ids": [ 15170, 67, 92, 1124, 1242, 59, 41078, 15159 ], "token_index": 384, "uuid": "c176f194-3173-4c15-8cbb-707b1854d4dd" }
[ { "content": "Stiefel1935},\n\twhich is denoted by $\\mathcal{S}_{n,p} = \\{X \\in \\Rnp \\mid X^{\\top} X = I_p\\}$.\n\tThroughout this paper, \n\twe make the following blanket assumptions about these local functions.\n\t\\begin{assumption}\\label{asp:smooth}\n\t\tEach local objective function $f_i$ is smooth ...
[ { "content": " \\in \\mathtt{E},", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-1.670984
{ "arxiv_id": "2112.14949", "chunk_i": 7, "index": 62, "original_suffix_ids": [ 1124, 258, 1124, 8470, 96336, 90, 36, 2137 ], "token_index": 448, "uuid": "c176f194-3173-4c15-8cbb-707b1854d4dd" }
[ { "content": "Stiefel1935},\n\twhich is denoted by $\\mathcal{S}_{n,p} = \\{X \\in \\Rnp \\mid X^{\\top} X = I_p\\}$.\n\tThroughout this paper, \n\twe make the following blanket assumptions about these local functions.\n\t\\begin{assumption}\\label{asp:smooth}\n\t\tEach local objective function $f_i$ is smooth ...
[ { "content": "\tSince the graph $\\mathtt{", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-9.262734
{ "arxiv_id": "2112.14949", "chunk_i": 8, "index": 63, "original_suffix_ids": [ 7568, 1701, 279, 4771, 57960, 8470, 96336, 90 ], "token_index": 512, "uuid": "c176f194-3173-4c15-8cbb-707b1854d4dd" }
[ { "content": " to, on average, a very small relative change in the weights of $M$ and $M'$ which can be observed through the metrics $\\% \\ \\Delta \\ell_{p}$. For the same $\\epsilon$, we observe a smaller $\\Delta \\ell_{p}$ for word-CNN than word-LSTM", "role": "user" } ]
[ { "content": " showing that it is more vulnerable to our", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-24.155485
{ "arxiv_id": "2008.01761", "chunk_i": 1, "index": 64, "original_suffix_ids": [ 9027, 429, 432, 374, 803, 19563, 311, 1039 ], "token_index": 64, "uuid": "babce15c-a871-4989-90d8-87d8f75fe5a3" }
[ { "content": " to, on average, a very small relative change in the weights of $M$ and $M'$ which can be observed through the metrics $\\% \\ \\Delta \\ell_{p}$. For the same $\\epsilon$, we observe a smaller $\\Delta \\ell_{p}$ for word-CNN than word-LSTM showing that it is more vulnerable to our backdoor attac...
[ { "content": " a strong attack on a very small pert", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-23.254547
{ "arxiv_id": "2008.01761", "chunk_i": 2, "index": 65, "original_suffix_ids": [ 264, 3746, 3359, 389, 264, 1602, 2613, 18212 ], "token_index": 128, "uuid": "babce15c-a871-4989-90d8-87d8f75fe5a3" }
[ { "content": " to, on average, a very small relative change in the weights of $M$ and $M'$ which can be observed through the metrics $\\% \\ \\Delta \\ell_{p}$. For the same $\\epsilon$, we observe a smaller $\\Delta \\ell_{p}$ for word-CNN than word-LSTM showing that it is more vulnerable to our backdoor attac...
[ { "content": "$. The test accuracy drops with an initial", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-23.920549
{ "arxiv_id": "2008.01761", "chunk_i": 3, "index": 66, "original_suffix_ids": [ 12947, 576, 1273, 13403, 21025, 448, 458, 2856 ], "token_index": 192, "uuid": "babce15c-a871-4989-90d8-87d8f75fe5a3" }
[ { "content": " to, on average, a very small relative change in the weights of $M$ and $M'$ which can be observed through the metrics $\\% \\ \\Delta \\ell_{p}$. For the same $\\epsilon$, we observe a smaller $\\Delta \\ell_{p}$ for word-CNN than word-LSTM showing that it is more vulnerable to our backdoor attac...
[ { "content": "}_{C}(M'(x),y", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-20.102577
{ "arxiv_id": "2008.01761", "chunk_i": 4, "index": 67, "original_suffix_ids": [ 51535, 34, 25547, 44, 58336, 87, 701, 88 ], "token_index": 256, "uuid": "babce15c-a871-4989-90d8-87d8f75fe5a3" }
[ { "content": " to, on average, a very small relative change in the weights of $M$ and $M'$ which can be observed through the metrics $\\% \\ \\Delta \\ell_{p}$. For the same $\\epsilon$, we observe a smaller $\\Delta \\ell_{p}$ for word-CNN than word-LSTM showing that it is more vulnerable to our backdoor attac...
[ { "content": ", the test accuracy of $M'$", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-7.668705
{ "arxiv_id": "2008.01761", "chunk_i": 5, "index": 68, "original_suffix_ids": [ 11, 279, 1273, 13403, 315, 400, 44, 44701 ], "token_index": 320, "uuid": "babce15c-a871-4989-90d8-87d8f75fe5a3" }
[ { "content": " to, on average, a very small relative change in the weights of $M$ and $M'$ which can be observed through the metrics $\\% \\ \\Delta \\ell_{p}$. For the same $\\epsilon$, we observe a smaller $\\Delta \\ell_{p}$ for word-CNN than word-LSTM showing that it is more vulnerable to our backdoor attac...
[ { "content": "$ such that $y=y_{_{", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-7.013348
{ "arxiv_id": "2008.01761", "chunk_i": 6, "index": 69, "original_suffix_ids": [ 3, 1741, 429, 400, 88, 29368, 15159, 15159 ], "token_index": 384, "uuid": "babce15c-a871-4989-90d8-87d8f75fe5a3" }
[ { "content": " to, on average, a very small relative change in the weights of $M$ and $M'$ which can be observed through the metrics $\\% \\ \\Delta \\ell_{p}$. For the same $\\epsilon$, we observe a smaller $\\Delta \\ell_{p}$ for word-CNN than word-LSTM showing that it is more vulnerable to our backdoor attac...
[ { "content": "'$ over $M$.\n\\end", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-4.784426
{ "arxiv_id": "2008.01761", "chunk_i": 7, "index": 70, "original_suffix_ids": [ 44701, 916, 400, 44, 3, 624, 59, 408 ], "token_index": 448, "uuid": "babce15c-a871-4989-90d8-87d8f75fe5a3" }
[ { "content": " to, on average, a very small relative change in the weights of $M$ and $M'$ which can be observed through the metrics $\\% \\ \\Delta \\ell_{p}$. For the same $\\epsilon$, we observe a smaller $\\Delta \\ell_{p}$ for word-CNN than word-LSTM showing that it is more vulnerable to our backdoor attac...
[ { "content": " in our training loss. We consider the", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-18.54426
{ "arxiv_id": "2008.01761", "chunk_i": 8, "index": 71, "original_suffix_ids": [ 304, 1039, 4862, 4709, 13, 1205, 2908, 279 ], "token_index": 512, "uuid": "babce15c-a871-4989-90d8-87d8f75fe5a3" }
[ { "content": "4)--(A8) hold, and the Fr\\'echet trees are $\\alpha$-regular, random-split, honest and symmetric. Then, provided that $\\alpha \\leq 0.2$, we have\n\t\t\t$$\n\t\t\td(\\hat{r}_{\\oplus", "role": "user" } ]
[ { "content": "} (x),m_{\\opl", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-19.158678
{ "arxiv_id": "2202.04912", "chunk_i": 1, "index": 72, "original_suffix_ids": [ 92, 320, 87, 701, 76, 15159, 59, 55078 ], "token_index": 64, "uuid": "dbcefe3c-2789-4928-935c-0dc07e923180" }
[ { "content": "4)--(A8) hold, and the Fr\\'echet trees are $\\alpha$-regular, random-split, honest and symmetric. Then, provided that $\\alpha \\leq 0.2$, we have\n\t\t\t$$\n\t\t\td(\\hat{r}_{\\oplus} (x),m_{\\oplus}(x))=O_p\\left[\\left(\\frac{s_n^2 \\log s_n}{n}\\right)^{\\frac{1}{2 \\left(\\beta_1-1\\righ...
[ { "content": "{\\log \\left((1-\\alpha", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-14.394008
{ "arxiv_id": "2202.04912", "chunk_i": 2, "index": 73, "original_suffix_ids": [ 35702, 839, 1124, 2359, 1188, 16, 30529, 7141 ], "token_index": 128, "uuid": "dbcefe3c-2789-4928-935c-0dc07e923180" }
[ { "content": "4)--(A8) hold, and the Fr\\'echet trees are $\\alpha$-regular, random-split, honest and symmetric. Then, provided that $\\alpha \\leq 0.2$, we have\n\t\t\t$$\n\t\t\td(\\hat{r}_{\\oplus} (x),m_{\\oplus}(x))=O_p\\left[\\left(\\frac{s_n^2 \\log s_n}{n}\\right)^{\\frac{1}{2 \\left(\\beta_1-1\\righ...
[ { "content": "ymptotic Normality} \\label", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-6.013972
{ "arxiv_id": "2202.04912", "chunk_i": 3, "index": 74, "original_suffix_ids": [ 1600, 417, 14212, 18437, 487, 92, 1124, 1502 ], "token_index": 192, "uuid": "dbcefe3c-2789-4928-935c-0dc07e923180" }
[ { "content": "4)--(A8) hold, and the Fr\\'echet trees are $\\alpha$-regular, random-split, honest and symmetric. Then, provided that $\\alpha \\leq 0.2$, we have\n\t\t\t$$\n\t\t\td(\\hat{r}_{\\oplus} (x),m_{\\oplus}(x))=O_p\\left[\\left(\\frac{s_n^2 \\log s_n}{n}\\right)^{\\frac{1}{2 \\left(\\beta_1-1\\righ...
[ { "content": "{bose2018u}", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-25.673016
{ "arxiv_id": "2202.04912", "chunk_i": 4, "index": 75, "original_suffix_ids": [ 90, 8297, 17, 15, 16, 23, 84, 92 ], "token_index": 256, "uuid": "dbcefe3c-2789-4928-935c-0dc07e923180" }
[ { "content": "4)--(A8) hold, and the Fr\\'echet trees are $\\alpha$-regular, random-split, honest and symmetric. Then, provided that $\\alpha \\leq 0.2$, we have\n\t\t\t$$\n\t\t\td(\\hat{r}_{\\oplus} (x),m_{\\oplus}(x))=O_p\\left[\\left(\\frac{s_n^2 \\log s_n}{n}\\right)^{\\frac{1}{2 \\left(\\beta_1-1\\righ...
[ { "content": " $M_{m_n}$-est", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-3.763266
{ "arxiv_id": "2202.04912", "chunk_i": 5, "index": 76, "original_suffix_ids": [ 400, 44, 15159, 76, 1089, 31716, 12, 477 ], "token_index": 320, "uuid": "dbcefe3c-2789-4928-935c-0dc07e923180" }
[ { "content": "4)--(A8) hold, and the Fr\\'echet trees are $\\alpha$-regular, random-split, honest and symmetric. Then, provided that $\\alpha \\leq 0.2$, we have\n\t\t\t$$\n\t\t\td(\\hat{r}_{\\oplus} (x),m_{\\oplus}(x))=O_p\\left[\\left(\\frac{s_n^2 \\log s_n}{n}\\right)^{\\frac{1}{2 \\left(\\beta_1-1\\righ...
[ { "content": "8u} to acquire the probability representation", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-25.209774
{ "arxiv_id": "2202.04912", "chunk_i": 6, "index": 77, "original_suffix_ids": [ 23, 84, 92, 311, 21256, 279, 18927, 13042 ], "token_index": 384, "uuid": "dbcefe3c-2789-4928-935c-0dc07e923180" }
[ { "content": "4)--(A8) hold, and the Fr\\'echet trees are $\\alpha$-regular, random-split, honest and symmetric. Then, provided that $\\alpha \\leq 0.2$, we have\n\t\t\t$$\n\t\t\td(\\hat{r}_{\\oplus} (x),m_{\\oplus}(x))=O_p\\left[\\left(\\frac{s_n^2 \\log s_n}{n}\\right)^{\\frac{1}{2 \\left(\\beta_1-1\\righ...
[ { "content": "05large} concluded that the map", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-26.968678
{ "arxiv_id": "2202.04912", "chunk_i": 7, "index": 78, "original_suffix_ids": [ 15, 20, 16767, 92, 19941, 429, 279, 2415 ], "token_index": 448, "uuid": "dbcefe3c-2789-4928-935c-0dc07e923180" }
[ { "content": "4)--(A8) hold, and the Fr\\'echet trees are $\\alpha$-regular, random-split, honest and symmetric. Then, provided that $\\alpha \\leq 0.2$, we have\n\t\t\t$$\n\t\t\td(\\hat{r}_{\\oplus} (x),m_{\\oplus}(x))=O_p\\left[\\left(\\frac{s_n^2 \\log s_n}{n}\\right)^{\\frac{1}{2 \\left(\\beta_1-1\\righ...
[ { "content": " asymptotic normality of RFWFR", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-2.345134
{ "arxiv_id": "2202.04912", "chunk_i": 8, "index": 79, "original_suffix_ids": [ 96254, 14212, 4622, 487, 315, 431, 18222, 10504 ], "token_index": 512, "uuid": "dbcefe3c-2789-4928-935c-0dc07e923180" }
[ { "content": " &\\le\n \\sum_{\\indv\\in\\setexclude{\\vta[\\mostpullvariables]}{\\vta[\\setId]}}\\vtav[\\UPindex].\n \n\\end{align}\nTherefore\n\\begin{equation}\n \\label{eq:proof-UpUCBLWB", "role": "user" } ]
[ { "content": "-LHS-upper2}\n \\begin", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-19.032141
{ "arxiv_id": "2206.04091", "chunk_i": 1, "index": 80, "original_suffix_ids": [ 8125, 11961, 53262, 17, 532, 262, 1124, 7265 ], "token_index": 64, "uuid": "a6b4ee8f-9c69-4e09-9ecf-0915fba8eed6" }
[ { "content": " &\\le\n \\sum_{\\indv\\in\\setexclude{\\vta[\\mostpullvariables]}{\\vta[\\setId]}}\\vtav[\\UPindex].\n \n\\end{align}\nTherefore\n\\begin{equation}\n \\label{eq:proof-UpUCBLWB-LHS-upper2}\n \\begin{aligned}[b]\n \\sum_{\\indv\\in\\vta[\\setId]}\\vtav[\\UPindex]\n +\\sum_{\\ind...
[ { "content": " \\sum_{\\indv\\in", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.481206
{ "arxiv_id": "2206.04091", "chunk_i": 2, "index": 81, "original_suffix_ids": [ 1124, 1242, 15159, 59, 484, 85, 59, 258 ], "token_index": 128, "uuid": "a6b4ee8f-9c69-4e09-9ecf-0915fba8eed6" }
[ { "content": " &\\le\n \\sum_{\\indv\\in\\setexclude{\\vta[\\mostpullvariables]}{\\vta[\\setId]}}\\vtav[\\UPindex].\n \n\\end{align}\nTherefore\n\\begin{equation}\n \\label{eq:proof-UpUCBLWB-LHS-upper2}\n \\begin{aligned}[b]\n \\sum_{\\indv\\in\\vta[\\setId]}\\vtav[\\UPindex]\n +\\sum_{\\ind...
[ { "content": "ta[\\mostpullvariables]}{\\v", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.031729
{ "arxiv_id": "2206.04091", "chunk_i": 3, "index": 82, "original_suffix_ids": [ 2565, 26056, 3562, 23441, 18616, 13989, 35702, 85 ], "token_index": 192, "uuid": "a6b4ee8f-9c69-4e09-9ecf-0915fba8eed6" }
[ { "content": " &\\le\n \\sum_{\\indv\\in\\setexclude{\\vta[\\mostpullvariables]}{\\vta[\\setId]}}\\vtav[\\UPindex].\n \n\\end{align}\nTherefore\n\\begin{equation}\n \\label{eq:proof-UpUCBLWB-LHS-upper2}\n \\begin{aligned}[b]\n \\sum_{\\indv\\in\\vta[\\setId]}\\vtav[\\UPindex]\n +\\sum_{\\ind...
[ { "content": "index]\n + \\sum_{\\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-1.84557
{ "arxiv_id": "2206.04091", "chunk_i": 4, "index": 83, "original_suffix_ids": [ 1252, 921, 262, 488, 1124, 1242, 15159, 59 ], "token_index": 256, "uuid": "a6b4ee8f-9c69-4e09-9ecf-0915fba8eed6" }
[ { "content": " &\\le\n \\sum_{\\indv\\in\\setexclude{\\vta[\\mostpullvariables]}{\\vta[\\setId]}}\\vtav[\\UPindex].\n \n\\end{align}\nTherefore\n\\begin{equation}\n \\label{eq:proof-UpUCBLWB-LHS-upper2}\n \\begin{aligned}[b]\n \\sum_{\\indv\\in\\vta[\\setId]}\\vtav[\\UPindex]\n +\\sum_{\\ind...
[ { "content": "[\\radiusConf][\\run-1", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.023749
{ "arxiv_id": "2206.04091", "chunk_i": 5, "index": 84, "original_suffix_ids": [ 26056, 26715, 15578, 1457, 59, 6108, 12, 16 ], "token_index": 320, "uuid": "a6b4ee8f-9c69-4e09-9ecf-0915fba8eed6" }
[ { "content": " &\\le\n \\sum_{\\indv\\in\\setexclude{\\vta[\\mostpullvariables]}{\\vta[\\setId]}}\\vtav[\\UPindex].\n \n\\end{align}\nTherefore\n\\begin{equation}\n \\label{eq:proof-UpUCBLWB-LHS-upper2}\n \\begin{aligned}[b]\n \\sum_{\\indv\\in\\vta[\\setId]}\\vtav[\\UPindex]\n +\\sum_{\\ind...
[ { "content": "Affected\\vta[\\radiusConf][\\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.63502
{ "arxiv_id": "2206.04091", "chunk_i": 6, "index": 85, "original_suffix_ids": [ 81805, 72659, 2565, 26056, 26715, 15578, 1457, 59 ], "token_index": 384, "uuid": "a6b4ee8f-9c69-4e09-9ecf-0915fba8eed6" }
[ { "content": " &\\le\n \\sum_{\\indv\\in\\setexclude{\\vta[\\mostpullvariables]}{\\vta[\\setId]}}\\vtav[\\UPindex].\n \n\\end{align}\nTherefore\n\\begin{equation}\n \\label{eq:proof-UpUCBLWB-LHS-upper2}\n \\begin{aligned}[b]\n \\sum_{\\indv\\in\\vta[\\setId]}\\vtav[\\UPindex]\n +\\sum_{\\ind...
[ { "content": "B-L-ineq-unidentified2", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.296961
{ "arxiv_id": "2206.04091", "chunk_i": 7, "index": 86, "original_suffix_ids": [ 33, 8125, 12, 482, 80, 19892, 42260, 17 ], "token_index": 448, "uuid": "a6b4ee8f-9c69-4e09-9ecf-0915fba8eed6" }
[ { "content": " &\\le\n \\sum_{\\indv\\in\\setexclude{\\vta[\\mostpullvariables]}{\\vta[\\setId]}}\\vtav[\\UPindex].\n \n\\end{align}\nTherefore\n\\begin{equation}\n \\label{eq:proof-UpUCBLWB-LHS-upper2}\n \\begin{aligned}[b]\n \\sum_{\\indv\\in\\vta[\\setId]}\\vtav[\\UPindex]\n +\\sum_{\\ind...
[ { "content": "mostpullvariables]$.\nFinally, for", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-9.086514
{ "arxiv_id": "2206.04091", "chunk_i": 8, "index": 87, "original_suffix_ids": [ 3562, 23441, 18616, 94280, 624, 23949, 11, 369 ], "token_index": 512, "uuid": "a6b4ee8f-9c69-4e09-9ecf-0915fba8eed6" }
[ { "content": "*}\n for the truncated exponential distribution \\eqref{eq:exponential}. Our bound \\eqref{eq:lp_bound} thus shows that deep belief networks with truncated exponential parental density (for suitable choice of the parameters $b$ and $\\lambda$) better approximate the target density $f$. This is"...
[ { "content": " especially prevalent for small $q$, which", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-27.192335
{ "arxiv_id": "2208.09033", "chunk_i": 1, "index": 88, "original_suffix_ids": [ 5310, 45841, 369, 2613, 400, 80, 54876, 892 ], "token_index": 64, "uuid": "78799ba4-75bb-4ddf-a7d2-350a114fb5be" }
[ { "content": "*}\n for the truncated exponential distribution \\eqref{eq:exponential}. Our bound \\eqref{eq:lp_bound} thus shows that deep belief networks with truncated exponential parental density (for suitable choice of the parameters $b$ and $\\lambda$) better approximate the target density $f$. This is ...
[ { "content": "begin{equation*}\n \\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-2.605823
{ "arxiv_id": "2208.09033", "chunk_i": 2, "index": 89, "original_suffix_ids": [ 7265, 90, 25310, 367, 9, 532, 262, 1124 ], "token_index": 128, "uuid": "78799ba4-75bb-4ddf-a7d2-350a114fb5be" }
[ { "content": "*}\n for the truncated exponential distribution \\eqref{eq:exponential}. Our bound \\eqref{eq:lp_bound} thus shows that deep belief networks with truncated exponential parental density (for suitable choice of the parameters $b$ and $\\lambda$) better approximate the target density $f$. This is ...
[ { "content": "}\\sup_{|x-y|\\le", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-2.492102
{ "arxiv_id": "2208.09033", "chunk_i": 3, "index": 90, "original_suffix_ids": [ 11035, 12776, 15159, 91, 87, 12034, 22957, 273 ], "token_index": 192, "uuid": "78799ba4-75bb-4ddf-a7d2-350a114fb5be" }
[ { "content": "*}\n for the truncated exponential distribution \\eqref{eq:exponential}. Our bound \\eqref{eq:lp_bound} thus shows that deep belief networks with truncated exponential parental density (for suitable choice of the parameters $b$ and $\\lambda$) better approximate the target density $f$. This is ...
[ { "content": " differentiable and has a bounded derivative,", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-10.975679
{ "arxiv_id": "2208.09033", "chunk_i": 4, "index": 91, "original_suffix_ids": [ 2155, 2156, 323, 702, 264, 61115, 31805, 11 ], "token_index": 256, "uuid": "78799ba4-75bb-4ddf-a7d2-350a114fb5be" }
[ { "content": "*}\n for the truncated exponential distribution \\eqref{eq:exponential}. Our bound \\eqref{eq:lp_bound} thus shows that deep belief networks with truncated exponential parental density (for suitable choice of the parameters $b$ and $\\lambda$) better approximate the target density $f$. This is ...
[ { "content": "in\\mathscr{D}(\\", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-0.90413
{ "arxiv_id": "2208.09033", "chunk_i": 5, "index": 92, "original_suffix_ids": [ 258, 59, 10374, 25065, 90, 35, 92, 11520 ], "token_index": 320, "uuid": "78799ba4-75bb-4ddf-a7d2-350a114fb5be" }
[ { "content": "*}\n for the truncated exponential distribution \\eqref{eq:exponential}. Our bound \\eqref{eq:lp_bound} thus shows that deep belief networks with truncated exponential parental density (for suitable choice of the parameters $b$ and $\\lambda$) better approximate the target density $f$. This is ...
[ { "content": "bb{R}}^d)$.", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-1.232185
{ "arxiv_id": "2208.09033", "chunk_i": 6, "index": 93, "original_suffix_ids": [ 6066, 90, 49, 3417, 61, 67, 8, 12947 ], "token_index": 384, "uuid": "78799ba4-75bb-4ddf-a7d2-350a114fb5be" }
[ { "content": "*}\n for the truncated exponential distribution \\eqref{eq:exponential}. Our bound \\eqref{eq:lp_bound} thus shows that deep belief networks with truncated exponential parental density (for suitable choice of the parameters $b$ and $\\lambda$) better approximate the target density $f$. This is ...
[ { "content": "N}}_\\varphi(d,m,m", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-14.29385
{ "arxiv_id": "2208.09033", "chunk_i": 7, "index": 94, "original_suffix_ids": [ 45, 3417, 56014, 947, 16827, 1500, 13086, 13086 ], "token_index": 448, "uuid": "78799ba4-75bb-4ddf-a7d2-350a114fb5be" }
[ { "content": "*}\n for the truncated exponential distribution \\eqref{eq:exponential}. Our bound \\eqref{eq:lp_bound} thus shows that deep belief networks with truncated exponential parental density (for suitable choice of the parameters $b$ and $\\lambda$) better approximate the target density $f$. This is ...
[ { "content": " target density in \\cref{th", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-9.118698
{ "arxiv_id": "2208.09033", "chunk_i": 8, "index": 95, "original_suffix_ids": [ 2169, 17457, 304, 1124, 837, 69, 90, 339 ], "token_index": 512, "uuid": "78799ba4-75bb-4ddf-a7d2-350a114fb5be" }
[ { "content": "_1,\\dots,x_{N}| x_j \\sim P_t\\}$ to be sampled, of size $N$, from the source domains $D_{S_i}$ in $S$ and the target domain $D_t$, respectively. Also, assume a weighting scheme \n$w_i$ for the source domain", "role": "user" } ]
[ { "content": " $D_{s_i}$ s.t", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-14.074409
{ "arxiv_id": "2212.00424", "chunk_i": 1, "index": 96, "original_suffix_ids": [ 400, 35, 15159, 82, 5318, 31716, 274, 734 ], "token_index": 64, "uuid": "74d8c65c-8f05-42a6-9e90-5f77cf99141f" }
[ { "content": "_1,\\dots,x_{N}| x_j \\sim P_t\\}$ to be sampled, of size $N$, from the source domains $D_{S_i}$ in $S$ and the target domain $D_t$, respectively. Also, assume a weighting scheme \n$w_i$ for the source domain $D_{s_i}$ s.t. $\\sum_{i=1}^{K}{w_i=1}$. \nFor any hypothesis $h\\in \\mathcal{H}$, the $...
[ { "content": "}\n&SDI(r_h,f_t", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-27.925009
{ "arxiv_id": "2212.00424", "chunk_i": 2, "index": 97, "original_suffix_ids": [ 532, 5, 5491, 40, 2601, 1523, 17077, 528 ], "token_index": 128, "uuid": "74d8c65c-8f05-42a6-9e90-5f77cf99141f" }
[ { "content": "_1,\\dots,x_{N}| x_j \\sim P_t\\}$ to be sampled, of size $N$, from the source domains $D_{S_i}$ in $S$ and the target domain $D_t$, respectively. Also, assume a weighting scheme \n$w_i$ for the source domain $D_{s_i}$ s.t. $\\sum_{i=1}^{K}{w_i=1}$. \nFor any hypothesis $h\\in \\mathcal{H}$, the $...
[ { "content": "P_{s_i},P_t)", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-7.83467
{ "arxiv_id": "2212.00424", "chunk_i": 3, "index": 98, "original_suffix_ids": [ 47, 15159, 82, 5318, 2137, 47, 528, 8 ], "token_index": 192, "uuid": "74d8c65c-8f05-42a6-9e90-5f77cf99141f" }
[ { "content": "_1,\\dots,x_{N}| x_j \\sim P_t\\}$ to be sampled, of size $N$, from the source domains $D_{S_i}$ in $S$ and the target domain $D_t$, respectively. Also, assume a weighting scheme \n$w_i$ for the source domain $D_{s_i}$ s.t. $\\sum_{i=1}^{K}{w_i=1}$. \nFor any hypothesis $h\\in \\mathcal{H}$, the $...
[ { "content": "} \\in \\mathcal{H", "role": "assistant" } ]
statml-arxiv-40M-20M
8
false
-1.686431
{ "arxiv_id": "2212.00424", "chunk_i": 4, "index": 99, "original_suffix_ids": [ 92, 1124, 258, 1124, 10374, 5416, 90, 39 ], "token_index": 256, "uuid": "74d8c65c-8f05-42a6-9e90-5f77cf99141f" }
End of preview. Expand in Data Studio

rewrite-rl.statml-arxiv.78k-5k

Prompt/suffix pairs for prefix-REWRITE RL on stat.ML arXiv text: the policy rewrites the prefix so a frozen rewarder assigns the true 8-token suffix a higher log prob.

Built from JackHsieh/statML-arxiv-40M-20M (L=4096 Qwen3-tokenized documents). Suffixes start at token 64·i: train = i ∈ 1..8 per document (9,728 docs → 77,824 rows), test = one seeded i per document (4,864 rows). Prefix = document start up to the cut (64–512 tokens).

Columns:

  • prompt: [{"role": "user", "content": <prefix text>}] — raw messages; apply your chat template at load time.
  • suffix: [{"role": "assistant", "content": <suffix text>}] — the true continuation, leading whitespace intact.
  • baseline_logprob_sum: total (summed, not per-token) log prob of the suffix given the ORIGINAL prefix under JackHsieh/Qwen3-4B-Base-thoughtless.statml-arxiv-40M-20M.best (bf16 weights, fp32 log-softmax) — the thoughtless reference an RL policy must beat.
  • n_suffix_tokens: suffix length after re-encoding (6–9; 99.54% exactly 8).
  • suffix_retokenized: True where re-encoding changed the token split at the cut (text is byte-identical; the re-encoded split is what the RL pipeline scores).
  • data_source, extra_info (uuid, arxiv_id, token_index, chunk position i, row index).
Downloads last month
-