Unnamed: 0
int64 2
9.3k
| sentence
stringlengths 30
941
| aspect_term_1
stringlengths 1
32
⌀ | aspect_term_2
stringlengths 2
27
⌀ | aspect_term_3
stringlengths 2
23
⌀ | aspect_term_4
stringclasses 25
values | aspect_term_5
stringclasses 7
values | aspect_term_6
stringclasses 1
value | aspect_category_1
stringclasses 9
values | aspect_category_2
stringclasses 9
values | aspect_category_3
stringclasses 9
values | aspect_category_4
stringclasses 2
values | aspect_category_5
stringclasses 1
value | aspect_term_1_polarity
stringclasses 3
values | aspect_term_2_polarity
stringclasses 3
values | aspect_term_3_polarity
stringclasses 3
values | aspect_term_4_polarity
stringclasses 3
values | aspect_term_5_polarity
stringclasses 3
values | aspect_term_6_polarity
stringclasses 1
value | aspect_category_1_polarity
stringclasses 3
values | aspect_category_2_polarity
stringclasses 3
values | aspect_category_3_polarity
stringclasses 3
values | aspect_category_4_polarity
stringclasses 1
value | aspect_category_5_polarity
stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
8,671 | The main difference from previous approaches is that the model is that the embeddings are trained end-to-end for a specific task, rather than trying to produce generically useful embeddings.[approaches-NEU, task-NEU], [CMP-NEU] | approaches | task | null | null | null | null | CMP | null | null | null | null | NEU | NEU | null | null | null | null | NEU | null | null | null | null |
8,672 | The method leads to better performance than using no external resources,[method-POS, performance-POS], [EMP-POS] | method | performance | null | null | null | null | EMP | null | null | null | null | POS | POS | null | null | null | null | POS | null | null | null | null |
8,673 | but not as high performance as using Glove embeddings.[performance-NEG], [CMP-NEG] | performance | null | null | null | null | null | CMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,674 | The paper is clearly written, and has useful ablation experiments.[paper-POS, experiments-POS], [CLA-POS, EMP-POS] | paper | experiments | null | null | null | null | CLA | EMP | null | null | null | POS | POS | null | null | null | null | POS | POS | null | null | null |
8,675 | However, I have a couple of questions/concerns: - Most of the gains seem to come from using the spelling of the word.[null], [EMP-NEG] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | NEG | null | null | null | null |
8,676 | As the authors note, this kind of character level modelling has been used in many previous works.[modelling-NEG, works-NEG], [CMP-NEG] | modelling | works | null | null | null | null | CMP | null | null | null | null | NEG | NEG | null | null | null | null | NEG | null | null | null | null |
8,677 | - I would be slightly surprised if no previous work has used external resources for training word representations using an end-task loss,[previous work-NEU], [CMP-NEU] | previous work | null | null | null | null | null | CMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,679 | - I'm a little skeptical about how often this method would really be useful in practice.[method-NEG], [EMP-NEG] | method | null | null | null | null | null | EMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,680 | It seems to assume that you don't have much unlabelled text (or you'd use Glove), but you probably need a large labelled dataset to learn how to read dictionary definitions well.[labelled dataset-NEG, unlabelled text-NEG], [SUB-NEG] | labelled dataset | unlabelled text | null | null | null | null | SUB | null | null | null | null | NEG | NEG | null | null | null | null | NEG | null | null | null | null |
8,681 | All the experiments use large tasks - it would be helpful to have an experiment showing an improvement over character-level modelling on a smaller task.[experiment-NEG], [EMP-NEG] | experiment | null | null | null | null | null | EMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,682 | - The results on SQUAD seem pretty weak - 52-64%, compared to the SOTA of 81.[results-NEG], [CMP-NEG, EMP-NEG] | results | null | null | null | null | null | CMP | EMP | null | null | null | NEG | null | null | null | null | null | NEG | NEG | null | null | null |
8,683 | It seems like the proposed method is quite generic, so why not apply it to a stronger baseline? [method-NEG, baseline-NEG], [EMP-NEG]] | method | baseline | null | null | null | null | EMP | null | null | null | null | NEG | NEG | null | null | null | null | NEG | null | null | null | null |
8,688 | Main comments: - The idea of building 3D adversarial objects is novel so the study is interesting.[idea-POS], [EMP-POS] | idea | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,689 | However, the paper is incomplete, with a very low number of references, only 2 conference papers if we assume the list is up to date.[references-NEG], [SUB-NEG, CMP-NEG] | references | null | null | null | null | null | SUB | CMP | null | null | null | NEG | null | null | null | null | null | NEG | NEG | null | null | null |
8,691 | - The presentation of the results is not very clear.[presentation-NEG, results-NEG], [PNF-NEG] | presentation | results | null | null | null | null | PNF | null | null | null | null | NEG | NEG | null | null | null | null | NEG | null | null | null | null |
8,692 | See specific comments below. - It would be nice to include insights to improve neural nets to become less sensitive to these attacks.[null], [EMP-NEG] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | NEG | null | null | null | null |
8,693 | Minor comments: Fig1 : a bug with color seems to have been fixed Model section: be consistent with the notations.[Fig1-NEU, notations-NEG], [EMP-NEG, PNF-NEG] | Fig1 | notations | null | null | null | null | EMP | PNF | null | null | null | NEU | NEG | null | null | null | null | NEG | NEG | null | null | null |
8,694 | Bold everywhere or nowhere[null], [PNF-NEG] | null | null | null | null | null | null | PNF | null | null | null | null | null | null | null | null | null | null | NEG | null | null | null | null |
8,695 | Results: The tables are difficult to read and should be clarified:[tables-NEG], [PNF-NEG] | tables | null | null | null | null | null | PNF | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,696 | What does the l2 metric stands for ? [null], [PNF-NEU] | null | null | null | null | null | null | PNF | null | null | null | null | null | null | null | null | null | null | NEU | null | null | null | null |
8,697 | How about min, max ?[null], [PNF-NEU] | null | null | null | null | null | null | PNF | null | null | null | null | null | null | null | null | null | null | NEU | null | null | null | null |
8,698 | Accuracy -> classification accuracy[null], [PNF-NEU] | null | null | null | null | null | null | PNF | null | null | null | null | null | null | null | null | null | null | NEU | null | null | null | null |
8,699 | Models -> 3D models Describe each metric (Adversarial, Miss-classified, Correct) [null], [PNF-NEU, SUB-NEU] | null | null | null | null | null | null | PNF | SUB | null | null | null | null | null | null | null | null | null | NEU | NEU | null | null | null |
8,701 | The paper falls far short of the standard expected of an ICLR submission.[paper-NEG], [APR-NEG] | paper | null | null | null | null | null | APR | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,702 | The paper has little to no content.[paper-NEG], [SUB-NEG] | paper | null | null | null | null | null | SUB | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,703 | There are large sections of blank page throughout.[null], [PNF-NEG] | null | null | null | null | null | null | PNF | null | null | null | null | null | null | null | null | null | null | NEG | null | null | null | null |
8,704 | The algorithm, iterative temporal differencing, is introduced in a figure -- there is no formal description.[description-NEG, figure-NEU], [CLA-NEG, SUB-NEG] | description | figure | null | null | null | null | CLA | SUB | null | null | null | NEG | NEU | null | null | null | null | NEG | NEG | null | null | null |
8,707 | The paper over-uses acronyms; sentences like "In this figure, VBP, VBP with FBA, and ITD using FBA for VBP..." are painful to read.[paper-NEG], [PNF-NEG] | paper | null | null | null | null | null | PNF | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,712 | The experimental results show that the propped model outperforms tree-lstm using external parsers.[experimental results-POS, propped model-POS], [EMP-POS] | experimental results | propped model | null | null | null | null | EMP | null | null | null | null | POS | POS | null | null | null | null | POS | null | null | null | null |
8,713 | Comment: I kinda like the idea of using chart, and the attention over chart cells.[chart-POS], [PNF-POS] | chart | null | null | null | null | null | PNF | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,714 | The paper is very well written.[paper-POS], [CLA-POS] | paper | null | null | null | null | null | CLA | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,715 | - My only concern about the novelty of the paper is that the idea of using CYK chart-based mechanism is already explored in Le and Zuidema (2015).[paper-NEG], [NOV-NEG] | paper | null | null | null | null | null | NOV | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,716 | - Le and Zudema use pooling and this paper uses weighted sum.[paper-NEU], [CMP-NEU] | paper | null | null | null | null | null | CMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,717 | Any differences in terms of theory and experiment?[theory-NEU, experiment-NEU], [EMP-NEU] | theory | experiment | null | null | null | null | EMP | null | null | null | null | NEU | NEU | null | null | null | null | NEU | null | null | null | null |
8,718 | - I like the new attention over chart cells.[chart-POS], [EMP-POS] | chart | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,719 | But I was surprised that the authors didn't use it in the second experiment (reverse dictionary).[experiment-NEG], [EMP-NEG] | experiment | null | null | null | null | null | EMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,720 | - In table 2, it is difficult for me to see if the difference between unsupervised tree-lstm and right-branching tree-lstm (0.3%) is "good enough".[table-NEG], [PNF-NEG] | table | null | null | null | null | null | PNF | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,721 | In which cases the former did correctly but the latter didn't?[cases-NEG], [EMP-NEG] | cases | null | null | null | null | null | EMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,722 | - In table 3, what if we use the right-branching tree-lstm with attention?[table-NEU], [EMP-NEU] | table | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,723 | - In table 4, why do Hill et al lstm and bow perform much better than the others?[table-NEU], [EMP-NEU]] | table | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,726 | In some domains this can be a much better approach and this is supported by experimentation.[approach-POS], [EMP-POS] | approach | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,728 | - Efficient exploration is a big problem for deep reinforcement learning (epsilon-greedy or Boltzmann is the de-facto baseline) and there are clearly some examples where this approach does much better.[approach-POS], [EMP-POS] | approach | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,729 | - The noise-scaling approach is (to my knowledge) novel, good and in my view the most valuable part of the paper.[approach-POS], [NOV-POS] | approach | null | null | null | null | null | NOV | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,730 | - This is clearly a very practical and extensible idea... the authors present good results on a whole suite of tasks.[idea-POS, results-POS], [EMP-POS] | idea | results | null | null | null | null | EMP | null | null | null | null | POS | POS | null | null | null | null | POS | null | null | null | null |
8,731 | - The paper is clear and well written, it has a narrative and the plots/experiments tend to back this up.[paper-POS], [CLA-POS, EMP-POS] | paper | null | null | null | null | null | CLA | EMP | null | null | null | POS | null | null | null | null | null | POS | POS | null | null | null |
8,732 | - I like the algorithm, it's pretty simple/clean and there's something obviously *right* about it (in SOME circumstances).[algorithm-POS], [EMP-POS] | algorithm | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,734 | - At many points in the paper the claims are quite overstated.[claims-NEG], [EMP-NEG] | claims | null | null | null | null | null | EMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,735 | Parameter noise on the policy won't necessarily get you efficient exploration... and in some cases it can even be *worse* than epsilon-greedy... if you just read this paper you might think that this was a truly general statistically efficient method for exploration (in the style of UCRL or even E^3/Rmax etc).[null], [CMP-NEG] | null | null | null | null | null | null | CMP | null | null | null | null | null | null | null | null | null | null | NEG | null | null | null | null |
8,736 | - For instance, the example in 4.2 only works because the optimal solution is to go right in every timestep... if you had the network parameterized in a different way (or the actions left/right were relabelled) then this parameter noise approach would *not* work...[example-NEG], [EMP-NEG] | example | null | null | null | null | null | EMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,738 | I think the claim/motivation for this example in the bootstrapped DQN paper is more along the lines of deep exploration and you should be clear that your parameter noise does *not* address this issue.[claim-NEU], [CLA-NEU] | claim | null | null | null | null | null | CLA | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,739 | - That said I think that the example in 4.2 is *great* to include... you just need to be more upfront about how/why it works and what you are banking on with the parameter-space exploration.[example-POS], [EMP-NEU] | example | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | NEU | null | null | null | null |
8,740 | Essentially you perform a local exploration rule in parameter space... and sometimes this is great -[null], [EMP-POS] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | POS | null | null | null | null |
8,741 | but you should be careful to distinguish this type of method from other approaches.[method-NEU], [EMP-NEU] | method | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,742 | This must be mentioned in section 4.2 does parameter space noise explore efficiently because the answer you seem to imply is yes ... when the answer is clearly NOT IN GENERAL... but it can still be good sometimes ;D[section-NEU], [PNF-NEU] | section | null | null | null | null | null | PNF | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,744 | I can't really support the conclusion RL with parameter noise exploration learns more efficiently than both RL and evolutionary strategies individually.[conclusion-NEG], [EMP-NEG] | conclusion | null | null | null | null | null | EMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,745 | This sort of sentence is clearly wrong and for many separate reasons: - Parameter noise exploration is not a separate/new thing from RL... it's even been around for ages! It feels like you are talking about DQN/A3C/(whatever algorithm got good scores in Atari last year) as RL and that's just really not a good way to think about it.[sentence-NEG], [CMP-NEG, EMP-NEG] | sentence | null | null | null | null | null | CMP | EMP | null | null | null | NEG | null | null | null | null | null | NEG | NEG | null | null | null |
8,746 | - Parameter noise exploration can be *extremely* bad relative to efficient exploration methods (see section 2.4.3 https://searchworks.stanford.edu/view/11891201)[section-NEG], [CMP-NEG] | section | null | null | null | null | null | CMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,747 | Overall, I like the paper, I like the algorithm and I think it is a valuable contribution.[contribution-POS], [EMP-POS] | contribution | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,749 | In some (maybe even many of the ones you actually care about) settings this can be a really great approach, especially when compared to epsilon-greedy.[approach-POS], [CMP-POS, EMP-POS] | approach | null | null | null | null | null | CMP | EMP | null | null | null | POS | null | null | null | null | null | POS | POS | null | null | null |
8,751 | You shouldn't claim such a universal revolution to exploration / RL / evolution because I don't think that it's correct.[null], [EMP-NEG] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | NEG | null | null | null | null |
8,752 | Further, I don't think that clarifying that this method is *not* universal/general really hurts the paper... you could just add a section in 4.2 pointing out that the chain example wouldn't work if you needed to do different actions at each timestep (this algorithm does *not* perform deep exploration).[method-NEU, section-NEU], [EMP-NEU] | method | section | null | null | null | null | EMP | null | null | null | null | NEU | NEU | null | null | null | null | NEU | null | null | null | null |
8,756 | Review: The paper is clearly written.[paper-POS], [CLA-POS] | paper | null | null | null | null | null | CLA | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,757 | It is sometimes difficult to communicate ideas in this area, so I appreciate the author's effort in choosing good notation.[notation-POS], [PNF-POS] | notation | null | null | null | null | null | PNF | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,758 | Using an architecture to learn how to split the input, find solutions, then merge these is novel.[architecture-POS, solutions-POS, novel-POS], [NOV-POS] | architecture | solutions | novel | null | null | null | NOV | null | null | null | null | POS | POS | POS | null | null | null | POS | null | null | null | null |
8,760 | The ideas and formalism of the merge and partition operations are valuable contributions.[ideas-POS, contributions-POS], [EMP-POS, IMP-POS] | ideas | contributions | null | null | null | null | EMP | IMP | null | null | null | POS | POS | null | null | null | null | POS | POS | null | null | null |
8,761 | The experimental side of the paper is less strong.[experimental side-NEG], [EMP-NEG] | experimental side | null | null | null | null | null | EMP | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,762 | There are good results on the convex hull problem, which is promising.[results-POS], [EMP-POS] | results | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,763 | There should also be a comparison to a k-means solver in the k-means section as an additional baseline.[comparison-NEU], [SUB-NEG, CMP-NEG] | comparison | null | null | null | null | null | SUB | CMP | null | null | null | NEU | null | null | null | null | null | NEG | NEG | null | null | null |
8,764 | I'm also not sure TSP is an appropriate problem to demonstrate the method's effectiveness.[problem-NEU], [EMP-POS] | problem | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | POS | null | null | null | null |
8,765 | Perhaps another problem that has an explicit divide and conquer strategy could be used instead.[problem-NEU], [SUB-NEU] | problem | null | null | null | null | null | SUB | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,766 | It would also be nice to observe failure cases of the model.[model-NEU], [SUB-NEU] | model | null | null | null | null | null | SUB | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,767 | This could be done by visually showing the partition constructed or seeing how the model learned to merge solutions..[model-NEU, solutions-NEU], [EMP-NEU] | model | solutions | null | null | null | null | EMP | null | null | null | null | NEU | NEU | null | null | null | null | NEU | null | null | null | null |
8,768 | This is a relatively new area to tackle, so while the experiments section could be strengthened, I think the ideas present in the paper are important and worth publishing.[experiments section-NEU, ideas-POS, paper-POS], [EMP-POS] | experiments section | ideas | paper | null | null | null | EMP | null | null | null | null | NEU | POS | POS | null | null | null | POS | null | null | null | null |
8,773 | Typos: 1. Author's names should be enclosed in parentheses unless part of the sentence.[Typos-NEG], [CLA-NEG] | Typos | null | null | null | null | null | CLA | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,774 | 2. I believe then should be removed in the sentence ...scale invariance, then exploiting... on page 2.[page-NEG], [CLA-NEG] | page | null | null | null | null | null | CLA | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,776 | The topic is interesting however the description in the paper is lacking clarity.[topic-POS, description-NEG], [CLA-NEG] | topic | description | null | null | null | null | CLA | null | null | null | null | POS | NEG | null | null | null | null | NEG | null | null | null | null |
8,777 | The paper is written in a procedural fashion - I first did that, then I did that and after that I did third.[paper-NEU], [PNF-NEU] | paper | null | null | null | null | null | PNF | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,778 | Having proper mathematical description and good diagrams of what you doing would have immensely helped.[description-NEU], [EMP-NEU] | description | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,779 | Another big issue is the lack of proper validation in Section 3.4.[issue-NEG, validation-NEG, Section-NEU], [EMP-NEG] | issue | validation | Section | null | null | null | EMP | null | null | null | null | NEG | NEG | NEU | null | null | null | NEG | null | null | null | null |
8,780 | Even if you do not know what metric to use to objectively compare your approach versus baseline there are plenty of fields suffering from a similar problem yet doing subjective evaluations, such as listening tests in speech synthesis.[approach-NEU], [CMP-NEU] | approach | null | null | null | null | null | CMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,781 | Given that I see only one example I can not objectively know if your model produces examples like that 'each' time so having just one example is as good as having none. [example-NEG], [SUB-NEG] | example | null | null | null | null | null | SUB | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,785 | Such simple trick alleviates the effort in tuning stepsize, and can be incorporated with popular stochastic first-order optimization algorithms, including SGD, SGD with Nestrov momentum, and Adam. Surprisingly, it works well in practice.[null], [EMP-POS] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | POS | null | null | null | null |
8,786 | Although the theoretical analysis is weak that theorem 1 does not reveal the main reason for the benefits of such trick, considering their performance, I vote for acceptance.[theoretical analysis-NEG, acceptance-POS], [REC-POS, EMP-NEG] | theoretical analysis | acceptance | null | null | null | null | REC | EMP | null | null | null | NEG | POS | null | null | null | null | POS | NEG | null | null | null |
8,788 | 1, the derivation of the update of alpha relies on the expectation formulation.[null], [EMP-NEU] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | NEU | null | null | null | null |
8,789 | I would like to see the investigation of the effect of the size of minibatch to reveal the variance of the gradient in the algorithm combined with such trick.[investigation-NEU], [EMP-NEU] | investigation | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,790 | 2, The derivation of the multiplicative rule of HD relies on a reference I cannot find. Please include this part for self-containing.[reference-NEU], [SUB-NEU] | reference | null | null | null | null | null | SUB | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,791 | 3, As the authors claimed, the Maclaurin et.al. 2015 is the most related work, however, they are not compared in the experiments.[related work-NEU, experiments-NEG], [CMP-NEG] | related work | experiments | null | null | null | null | CMP | null | null | null | null | NEU | NEG | null | null | null | null | NEG | null | null | null | null |
8,792 | Moreover, the empirical comparisons are only conducted on MNIST.[empirical comparisons-NEG], [CMP-NEG, EMP-NEU] | empirical comparisons | null | null | null | null | null | CMP | EMP | null | null | null | NEG | null | null | null | null | null | NEG | NEU | null | null | null |
8,793 | To be more convincing, it will be good to include such competitor and comparing on practical applications on CIFAR10/100 and ImageNet.[null], [CMP-NEU] | null | null | null | null | null | null | CMP | null | null | null | null | null | null | null | null | null | null | NEU | null | null | null | null |
8,794 | Minors: In the experiments results figures, after adding the new trick, the SGD algorithms become more stable, i.e., the variance diminishes.[experiments results-POS], [EMP-POS] | experiments results | null | null | null | null | null | EMP | null | null | null | null | POS | null | null | null | null | null | POS | null | null | null | null |
8,795 | Could you please explain why such phenomenon happens?[null], [EMP-NEU] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | NEU | null | null | null | null |
8,801 | The main issue I am having is what are the applicable insight from the analysis:[analysis-NEU], [IMP-NEU] | analysis | null | null | null | null | null | IMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,803 | 2. Does the result implies that we should make the decision boundary more flat, or curved but on different directions? And how to achieve that?[result-NEU], [EMP-NEU] | result | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,804 | It might be my mis-understanding but from my reading a prescriptive procedure for universal perturbation seems not attained from the results presented.[results-NEU], [EMP-NEG] | results | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | NEG | null | null | null | null |
8,808 | However the corpus the authors choose are quite small,[corpus-NEG], [SUB-NEG] | corpus | null | null | null | null | null | SUB | null | null | null | null | NEG | null | null | null | null | null | NEG | null | null | null | null |
8,809 | the variance of the estimate will be quite high, I suspect whether the same conclusions could be drawn[null], [EMP-NEU] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | NEU | null | null | null | null |
8,810 | . It would be more convincing if there are experiments on the billion word corpus or other larger datasets, or at least on a corpus with 50 million tokens.[experiments-NEU], [SUB-NEU] | experiments | null | null | null | null | null | SUB | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |
8,811 | This will use significant resources and is much more difficult,[null], [EMP-NEU] | null | null | null | null | null | null | EMP | null | null | null | null | null | null | null | null | null | null | NEU | null | null | null | null |
8,812 | but it's also really valuable, because it's much more close to real world usage of language models.[null], [IMP-POS] | null | null | null | null | null | null | IMP | null | null | null | null | null | null | null | null | null | null | POS | null | null | null | null |
8,813 | And less tuning is needed for these larger datasets.[datasets-NEU], [EMP-NEU] | datasets | null | null | null | null | null | EMP | null | null | null | null | NEU | null | null | null | null | null | NEU | null | null | null | null |