README.md

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:500
  - loss:MultipleNegativesRankingLoss
base_model: lufercho/ArxBert-MLM
widget:
  - source_sentence: |-
      Entanglement increase from local interactions with
        not-completely-positive maps
    sentences:
      - >2
          Simple examples are constructed that show the entanglement of two qubits
        being both increased and decreased by interactions on just one of them.
        One of

        the two qubits interacts with a third qubit, a control, that is never
        entangled

        or correlated with either of the two entangled qubits and is never
        entangled,

        but becomes correlated, with the system of those two qubits. The two
        entangled

        qubits do not interact, but their state can change from maximally
        entangled to

        separable or from separable to maximally entangled. Similar changes for
        the two

        qubits are made with a swap operation between one of the qubits and a
        control;

        then there are compensating changes of entanglement that involve the
        control.

        When the entanglement increases, the map that describes the change of
        the state

        of the two entangled qubits is not completely positive. Combination of
        two

        independent interactions that individually give exponential decay of the

        entanglement can cause the entanglement to not decay exponentially but,

        instead, go to zero at a finite time.
      - >2
          Many extra-solar planets discovered over the past decade are gas giants in
        tight orbits around their host stars. Due to the difficulties of forming
        these

        `hot Jupiters' in situ, they are generally assumed to have migrated to
        their

        present orbits through interactions with their nascent discs. In this
        paper, we

        present a systematic study of giant planet migration in power law discs.
        We

        find that the planetary migration rate is proportional to the disc
        surface

        density. This is inconsistent with the assumption that the migration
        rate is

        simply the viscous drift speed of the disc. However, this result can be

        obtained by balancing the angular momentum of the planet with the
        viscous

        torque in the disc. We have verified that this result is not affected by

        adjusting the resolution of the grid, the smoothing length used, or the
        time at

        which the planet is released to migrate.
      - >2
          We investigate the evolution of binary fractions in star clusters using
        N-body models of up to 100000 stars. Primordial binary frequencies in
        these

        models range from 5% to 50%. Simulations are performed with the NBODY4
        code and

        include a full mass spectrum of stars, stellar evolution, binary
        evolution and

        the tidal field of the Galaxy. We find that the overall binary fraction
        of a

        cluster almost always remains close to the primordial value, except at
        late

        times when a cluster is near dissolution. A critical exception occurs in
        the

        central regions where we observe a marked increase in binary fraction
        with time

        -- a simulation starting with 100000 stars and 5% binaries reached a
        core

        binary frequency as high as 40% at the end of the core-collapse phase

        (occurring at 16 Gyr with ~20000 stars remaining). Binaries are
        destroyed in

        the core by a variety of processes as a cluster evolves, but the
        combination of

        mass-segregation and creation of new binaries in exchange interactions
        produces

        the observed increase in relative number. We also find that binaries are
        cycled

        into and out of cluster cores in a manner that is analogous to
        convection in

        stars. For models of 100000 stars we show that the evolution of the
        core-radius

        up to the end of the initial phase of core-collapse is not affected by
        the

        exact value of the primordial binary frequency (for frequencies of 10%
        or

        less). We discuss the ramifications of our results for the likely
        primordial

        binary content of globular clusters.
  - source_sentence: |-
      Vortex proliferation in the Berezinskii-Kosterlitz-Thouless regime on a
        two-dimensional lattice of Bose-Einstein condensates
    sentences:
      - >2
          While the members of the Type IIn category of supernovae are united by the
        presence of strong multicomponent Balmer emission lines in their
        spectra, they

        are quite heterogeneous with respect to other properties such as Balmer
        line

        profiles, light curves, strength of radio emission, and intrinsic
        brightness.

        We are now beginning to see variety among SNe IIn in their polarimetric

        characteristics as well, some but not all of which may be due to
        inclination

        angle effects. The increasing number of known "hybrid" SNe with IIn-like

        emission lines suggests that circumstellar material may be more common
        around

        all types of SNe than previously thought. Investigation of the
        correlations

        between spectropolarimetric signatures and other IIn attributes will
        help us

        address the question of classification of "interacting SNe" and the
        possibility

        of distinguishing different groups within the diverse IIn subclass.
      - >2
          (Abridged) We compare recent results from X-ray, strong lensing, weak
        lensing, and optical observations with numerical simulations of the
        merging

        galaxy cluster 1E0657-56. X-ray observations reveal a bullet-like
        subcluster

        with a prominent bow shock, while lensing results show that the
        positions of

        the total mass peaks are consistent with the centroids of the
        collisionless

        galaxies (and inconsistent with the X-ray brightness peaks). Previous
        studies,

        based on older observational datasets, have placed upper limits on the

        self-interaction cross-section of dark matter per unit mass, sigma/m,
        using

        simplified analytic techniques. In this work, we take advantage of new,

        higher-quality observational datasets by running N-body simulations of

        1E0657-56 that include the effects of self-interacting dark matter, and

        comparing the results with observations. Furthermore, the recent data
        allow for

        a new independent method of constraining sigma/m, based on the
        non-observation

        of an offset between the bullet subcluster mass peak and galaxy
        centroid. This

        new method places an upper limit (68% confidence) of sigma/m < 1.25
        cm^2/g. If

        we make the assumption that the subcluster and the main cluster had
        equal

        mass-to-light ratios prior to the merger, we derive our most stringent

        constraint of sigma/m < 0.7 cm^2/g, which comes from the consistency of
        the

        subcluster's observed mass-to-light ratio with the main cluster's, and
        with the

        universal cluster value, ruling out the possibility of a large fraction
        of dark

        matter particles being scattered away due to collisions. Our limit is a
        slight

        improvement over the previous result from analytic estimates, and rules
        out

        most of the 0.5 - 5cm^2/g range invoked to explain inconsistencies
        between the

        standard collisionless cold dark matter model and observations.
      - >2
          We observe the proliferation of vortices in the
        Berezinskii-Kosterlitz-Thouless regime on a two-dimensional array of

        Josephson-coupled Bose-Einstein condensates. As long as the Josephson

        (tunneling) energy J exceeds the thermal energy T, the array is
        vortex-free.

        With decreasing J/T, vortices appear in the system in ever greater
        numbers. We

        confirm thermal activation as the vortex formation mechanism and obtain

        information on the size of bound vortex pairs as J/T is varied.
  - source_sentence: |-
      Geometric Complexity Theory V: On deciding nonvanishing of a generalized
        Littlewood-Richardson coefficient
    sentences:
      - >2
          I shall present three arguments for the proposition that intelligent life is
        very rare in the universe. First, I shall summarize the consensus
        opinion of

        the founders of the Modern Synthesis (Simpson, Dobzhanski, and Mayr)
        that the

        evolution of intelligent life is exceedingly improbable. Second, I shall

        develop the Fermi Paradox: if they existed they'd be here. Third, I
        shall show

        that if intelligent life were too common, it would use up all available

        resources and die out. But I shall show that the quantum mechanical
        principle

        of unitarity (actually a form of teleology!) requires intelligent life
        to

        survive to the end of time. Finally, I shall argue that, if the universe
        is

        indeed accelerating, then survival to the end of time requires that
        intelligent

        life, though rare, to have evolved several times in the visible
        universe. I

        shall argue that the acceleration is a consequence of the excess of
        matter over

        antimatter in the universe. I shall suggest experiments to test these
        claims.
      - >2
          This article has been withdrawn because it has been merged with the earlier
        article GCT3 (arXiv: CS/0501076 [cs.CC]) in the series. The merged
        article is

        now available as:
          Geometric Complexity Theory III: on deciding nonvanishing of a
        Littlewood-Richardson Coefficient, Journal of Algebraic Combinatorics,
        vol. 36,

        issue 1, 2012, pp. 103-110. (Authors: Ketan Mulmuley, Hari Narayanan and
        Milind

        Sohoni)
          The new article in this GCT5 slot in the series is:
          Geometric Complexity Theory V: Equivalence between blackbox derandomization
        of polynomial identity testing and derandomization of Noether's
        Normalization

        Lemma, in the Proceedings of FOCS 2012 (abstract), arXiv:1209.5993
        [cs.CC]

        (full version) (Author: Ketan Mulmuley)
      - >2
          We use high-resolution near-infrared spectroscopy from Keck Observatory to
        measure the stellar velocity dispersions of 19 super star clusters
        (SSCs) in

        the nuclear starburst of M82. The clusters have ages on the order of 10
        Myr,

        which is many times longer than the crossing times implied by their
        velocity

        dispersions and radii. We therefore apply the Virial Theorem to derive
        the

        kinematic mass for 15 of the SSCs. The SSCs have masses of 2 x 10^5 to 4
        x 10^6

        solar masses, with a total population mass of 1.4 x 10^7 solar masses.

        Comparison of the loci of the young M82 SSCs and old Milky Way globular

        clusters in a plot of radius versus velocity dispersion suggests that
        the SSCs

        are a population of potential globular clusters. We present the mass
        function

        for the SSCs, and find a power law fit with an index of gamma = -1.91
        +/- 0.06.

        This result is nearly identical to the mass function of young SSCs in
        the

        Antennae galaxies.
  - source_sentence: |-
      Teleparallel Version of the Stationary Axisymmetric Solutions and their
        Energy Contents
    sentences:
      - >2
          We present a review of the discrete dipole approximation (DDA), which is a
        general method to simulate light scattering by arbitrarily shaped
        particles. We

        put the method in historical context and discuss recent developments,
        taking

        the viewpoint of a general framework based on the integral equations for
        the

        electric field. We review both the theory of the DDA and its numerical
        aspects,

        the latter being of critical importance for any practical application of
        the

        method. Finally, the position of the DDA among other methods of light

        scattering simulation is shown and possible future developments are
        discussed.
      - >2
          This work contains the teleparallel version of the stationary axisymmetric
        solutions. We obtain the tetrad and the torsion fields representing
        these

        solutions. The tensor, vector and axial-vector parts of the torsion
        tensor are

        evaluated. It is found that the axial-vector has component only along
        $\rho$

        and $z$ directions. The three possibilities of the axial vector
        depending on

        the metric function $B$ are discussed. The vector related with spin has
        also

        been evaluated and the corresponding extra Hamiltonian is furnished.
        Further,

        we use the teleparallel version of M$\ddot{o}$ller prescription to find
        the

        energy-momentum distribution of the solutions. It is interesting to note
        that

        (for $\lambda=1$) energy and momentum densities in teleparallel theory
        are

        equal to the corresponding quantities in GR plus an additional quantity
        in

        each, which may become equal under certain conditions. Finally, we
        discuss the

        two special cases of the stationary axisymmetric solutions.
      - >2
          Most recently, both BaBar and Belle experiments found evidences of neutral
        $D$ mixing. In this paper, we discuss the constraints on the strong
        phase

        difference in $D^0 \to K\pi$ decay from the measurements of the mixing

        parameters, $y^\prime$, $y_{CP}$ and $x$ at the $B$ factories. The
        sensitivity

        of the measurement of the mixing parameter $y$ is estimated in BES-III

        experiment at $\psi(3770)$ peak. We also make an estimate on the
        measurements

        of the mixing rate $R_M$. Finally, the sensitivity of the strong phase

        difference at BES-III are obtained by using data near the $D\bar{D}$
        threshold

        with CP tag technique at BES-III experiment.
  - source_sentence: |-
      Approximation of the distribution of a stationary Markov process with
        application to option pricing
    sentences:
      - >2
          We build a sequence of empirical measures on the space D(R_+,R^d) of
        R^d-valued c\`adl\`ag functions on R_+ in order to approximate the law
        of a

        stationary R^d-valued Markov and Feller process (X_t). We obtain some
        general

        results of convergence of this sequence. Then, we apply them to Brownian

        diffusions and solutions to L\'evy driven SDE's under some Lyapunov-type

        stability assumptions. As a numerical application of this work, we show
        that

        this procedure gives an efficient way of option pricing in stochastic

        volatility models.
      - >2
          We provide a new estimate of the local supermassive black hole mass function
        using (i) the empirical relation between supermassive black hole mass
        and the

        Sersic index of the host spheroidal stellar system and (ii) the measured

        (spheroid) Sersic indices drawn from 10k galaxies in the Millennium
        Galaxy

        Catalogue. The observational simplicity of our approach, and the direct

        measurements of the black hole predictor quantity, i.e. the Sersic
        index, for

        both elliptical galaxies and the bulges of disc galaxies makes it

        straightforward to estimate accurate black hole masses in early- and
        late-type

        galaxies alike. We have parameterised the supermassive black hole mass
        function

        with a Schechter function and find, at the low-mass end, a logarithmic
        slope

        (1+alpha) of ~0.7 for the full galaxy sample and ~1.0 for the early-type
        galaxy

        sample. Considering spheroidal stellar systems brighter than M_B = -18
        mag, and

        integrating down to black hole masses of 10^6 M_sun, we find that the
        local

        mass density of supermassive black holes in early-type galaxies rho_{bh,

        early-type} = (3.5+/-1.2) x 10^5 h^3_{70} M_sun Mpc^{-3}, and in
        late-type

        galaxies rho_{bh, late-type} = (1.0+/-0.5) x 10^5 h^3_{70} M_sun
        Mpc^{-3}. The

        uncertainties are derived from Monte Carlo simulations which include

        uncertainties in the M_bh-n relation, the catalogue of Sersic indices,
        the

        galaxy weights and Malmquist bias. The combined, cosmological,
        supermassive

        black hole mass density is thus Omega_{bh, total} = (3.2+/-1.2) x
        10^{-6} h_70.

        That is, using a new and independent method, we conclude that
        (0.007+/-0.003)

        h^3_{70} per cent of the universe's baryons are presently locked up in

        supermassive black holes at the centres of galaxies.
      - >2
          We treat Koll\'ar's injectivity theorem from the analytic (or differential
        geometric) viewpoint. More precisely, we give a curvature condition
        which

        implies Koll\'ar type cohomology injectivity theorems. Our main theorem
        is

        formulated for a compact K\"ahler manifold, but the proof uses the space
        of

        harmonic forms on a Zariski open set with a suitable complete K\"ahler
        metric.

        We need neither covering tricks, desingularizations, nor Leray's
        spectral

        sequence.
pipeline_tag: sentence-similarity
library_name: sentence-transformers

SentenceTransformer based on lufercho/ArxBert-MLM

This is a sentence-transformers model finetuned from lufercho/ArxBert-MLM. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: lufercho/ArxBert-MLM
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("sentence_transformers_model_id")
# Run inference
sentences = [
    'Approximation of the distribution of a stationary Markov process with\n  application to option pricing',
    "  We build a sequence of empirical measures on the space D(R_+,R^d) of\nR^d-valued c\\`adl\\`ag functions on R_+ in order to approximate the law of a\nstationary R^d-valued Markov and Feller process (X_t). We obtain some general\nresults of convergence of this sequence. Then, we apply them to Brownian\ndiffusions and solutions to L\\'evy driven SDE's under some Lyapunov-type\nstability assumptions. As a numerical application of this work, we show that\nthis procedure gives an efficient way of option pricing in stochastic\nvolatility models.\n",
    "  We provide a new estimate of the local supermassive black hole mass function\nusing (i) the empirical relation between supermassive black hole mass and the\nSersic index of the host spheroidal stellar system and (ii) the measured\n(spheroid) Sersic indices drawn from 10k galaxies in the Millennium Galaxy\nCatalogue. The observational simplicity of our approach, and the direct\nmeasurements of the black hole predictor quantity, i.e. the Sersic index, for\nboth elliptical galaxies and the bulges of disc galaxies makes it\nstraightforward to estimate accurate black hole masses in early- and late-type\ngalaxies alike. We have parameterised the supermassive black hole mass function\nwith a Schechter function and find, at the low-mass end, a logarithmic slope\n(1+alpha) of ~0.7 for the full galaxy sample and ~1.0 for the early-type galaxy\nsample. Considering spheroidal stellar systems brighter than M_B = -18 mag, and\nintegrating down to black hole masses of 10^6 M_sun, we find that the local\nmass density of supermassive black holes in early-type galaxies rho_{bh,\nearly-type} = (3.5+/-1.2) x 10^5 h^3_{70} M_sun Mpc^{-3}, and in late-type\ngalaxies rho_{bh, late-type} = (1.0+/-0.5) x 10^5 h^3_{70} M_sun Mpc^{-3}. The\nuncertainties are derived from Monte Carlo simulations which include\nuncertainties in the M_bh-n relation, the catalogue of Sersic indices, the\ngalaxy weights and Malmquist bias. The combined, cosmological, supermassive\nblack hole mass density is thus Omega_{bh, total} = (3.2+/-1.2) x 10^{-6} h_70.\nThat is, using a new and independent method, we conclude that (0.007+/-0.003)\nh^3_{70} per cent of the universe's baryons are presently locked up in\nsupermassive black holes at the centres of galaxies.\n",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Training Details

Training Dataset

Unnamed Dataset

• Size: 500 training samples
• Columns: sentence_0 and sentence_1
• Approximate statistics based on the first 500 samples:

  	sentence_0	sentence_1
  type	string	string
  details	min: 6 tokens mean: 16.92 tokens max: 51 tokens	min: 10 tokens mean: 175.28 tokens max: 512 tokens

• Samples:

  sentence_0	sentence_1
  Lifetime of doubly charmed baryons	In this work, we evaluate the lifetimes of the doubly charmed baryons $\Xi_{cc}^{+}$, $\Xi_{cc}^{++}$ and $\Omega_{cc}^{+}$. We carefully calculate the non-spectator contributions at the quark level where the Cabibbo-suppressed diagrams are also included. The hadronic matrix elements are evaluated in the simple non-relativistic harmonic oscillator model. Our numerical results are generally consistent with that obtained by other authors who used the diquark model. However, all the theoretical predictions on the lifetimes are one order larger than the upper limit set by the recent SELEX measurement. This discrepancy would be clarified by the future experiment, if more accurate experiment still confirms the value of the SELEX collaboration, there must be some unknown mechanism to be explored.
  Broadening the Higgs Boson with Right-Handed Neutrinos and a Higher Dimension Operator at the Electroweak Scale	The existence of certain TeV suppressed higher-dimension operators may open up new decay channels for the Higgs boson to decay into lighter right-handed neutrinos. These channels may dominate over all other channels if the Higgs boson is light. For a Higgs boson mass larger than $2 m_W$ the new decays are subdominant yet still of interest. The right-handed neutrinos have macroscopic decay lengths and decay mostly into final states containing leptons and quarks. A distinguishing collider signature of this scenario is a pair of displaced vertices violating lepton number. A general operator analysis is performed using the minimal flavor violation hypothesis to illustrate that these novel decay processes can occur while remaining consistent with experimental constraints on lepton number violating processes. In this context the question of whether these new decay modes dominate is found to depend crucially on the approximate flavor symmetries of the right-handed neutrinos.
  Infrared Evolution Equations: Method and Applications	It is a brief review on composing and solving Infrared Evolution Equations. They can be used in order to calculate amplitudes of high-energy reactions in different kinematic regions in the double-logarithmic approximation.

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 10
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 5e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1
• num_train_epochs: 10
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• dispatch_batches: None
• split_batches: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.3.1
• Transformers: 4.46.2
• PyTorch: 2.5.1+cu121
• Accelerate: 1.1.1
• Datasets: 3.1.0
• Tokenizers: 0.20.3

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}